Nim Manual

"Complexity" seems to be a lot like "energy": you can transfer it from the end user to one/some of the other players, but the total amount seems to remain pretty much constant for a given task. -- Ran

About this document

Note: This document is a draft! Several of Nim's features may need more precise wording. This manual is constantly evolving into a proper specification.

This document describes the lexis, the syntax, and the semantics of Nim.

The language constructs are explained using an extended BNF, in which (a)* means 0 or more a's, a+ means 1 or more a's, and (a)? means an optional a. Parentheses may be used to group elements.

& is the lookahead operator; &a means that an a is expected but not consumed. It will be consumed in the following rule.

The |, / symbols are used to mark alternatives and have the lowest precedence. / is the ordered choice that requires the parser to try the alternatives in the given order. / is often used to ensure the grammar is not ambiguous.

Non-terminals start with a lowercase letter, abstract terminal symbols are in UPPERCASE. Verbatim terminal symbols (including keywords) are quoted with '. An example:

ifStmt = 'if' expr ':' stmts ('elif' expr ':' stmts)* ('else' stmts)?

The binary ^* operator is used as a shorthand for 0 or more occurrences separated by its second argument; likewise ^+ means 1 or more occurrences: a ^+ b is short for a (b a)* and a ^* b is short for (a (b a)*)?. Example:

arrayConstructor = '[' expr ^* ',' ']'

Other parts of Nim - like scoping rules or runtime semantics are only described in the, more easily comprehensible, informal manner for now.


A Nim program specifies a computation that acts on a memory consisting of components called locations. A variable is basically a name for a location. Each variable and location is of a certain type. The variable's type is called static type, the location's type is called dynamic type. If the static type is not the same as the dynamic type, it is a super-type or subtype of the dynamic type.

An identifier is a symbol declared as a name for a variable, type, procedure, etc. The region of the program over which a declaration applies is called the scope of the declaration. Scopes can be nested. The meaning of an identifier is determined by the smallest enclosing scope in which the identifier is declared unless overloading resolution rules suggest otherwise.

An expression specifies a computation that produces a value or location. Expressions that produce locations are called l-values. An l-value can denote either a location or the value the location contains, depending on the context. Expressions whose values can be determined statically are called constant expressions; they are never l-values.

A static error is an error that the implementation detects before program execution. Unless explicitly classified, an error is a static error.

A checked runtime error is an error that the implementation detects and reports at runtime. The method for reporting such errors is via raising exceptions or dying with a fatal error. However, the implementation provides a means to disable these runtime checks. See the section pragmas for details.

Whether a checked runtime error results in an exception or in a fatal error at runtime is implementation specific. Thus the following program is always invalid:

var a: array[0..1, char]
let i = 5
  a[i] = 'N'
except IndexError:
  echo "invalid index"

An unchecked runtime error is an error that is not guaranteed to be detected, and can cause the subsequent behavior of the computation to be arbitrary. Unchecked runtime errors cannot occur if only safe language features are used.

Lexical Analysis


All Nim source files are in the UTF-8 encoding (or its ASCII subset). Other encodings are not supported. Any of the standard platform line termination sequences can be used - the Unix form using ASCII LF (linefeed), the Windows form using the ASCII sequence CR LF (return followed by linefeed), or the old Macintosh form using the ASCII CR (return) character. All of these forms can be used equally, regardless of platform.


Nim's standard grammar describes an indentation sensitive language. This means that all the control structures are recognized by indentation. Indentation consists only of spaces; tabulators are not allowed.

The indentation handling is implemented as follows: The lexer annotates the following token with the preceding number of spaces; indentation is not a separate token. This trick allows parsing of Nim with only 1 token of lookahead.

The parser uses a stack of indentation levels: the stack consists of integers counting the spaces. The indentation information is queried at strategic places in the parser but ignored otherwise: The pseudo terminal IND{>} denotes an indentation that consists of more spaces than the entry at the top of the stack; IND{=} an indentation that has the same number of spaces. DED is another pseudo terminal that describes the action of popping a value from the stack, IND{>} then implies to push onto the stack.

With this notation we can now easily define the core of the grammar: A block of statements (simplified example):

ifStmt = 'if' expr ':' stmt
         (IND{=} 'elif' expr ':' stmt)*
         (IND{=} 'else' ':' stmt)?

simpleStmt = ifStmt / ...

stmt = IND{>} stmt ^+ IND{=} DED  # list of statements
     / simpleStmt                 # or a simple statement


Comments start anywhere outside a string or character literal with the hash character #. Comments consist of a concatenation of comment pieces. A comment piece starts with # and runs until the end of the line. The end of line characters belong to the piece. If the next line only consists of a comment piece with no other tokens between it and the preceding one, it does not start a new comment:

i = 0     # This is a single comment over multiple lines.
  # The scanner merges these two pieces.
  # The comment continues here.

Documentation comments are comments that start with two ##. Documentation comments are tokens; they are only allowed at certain places in the input file as they belong to the syntax tree!

Multiline comments

Starting with version 0.13.0 of the language Nim supports multiline comments. They look like:

#[Comment here.
Multiple lines
are not a problem.]#

Multiline comments support nesting:

#[  #[ Multiline comment in already
   commented out code. ]#
proc p[T](x: T) = discard

Multiline documentation comments also exist and support nesting too:

proc foo =
  ##[Long documentation comment

Identifiers & Keywords

Identifiers in Nim can be any string of letters, digits and underscores, beginning with a letter. Two immediate following underscores __ are not allowed:

letter ::= 'A'..'Z' | 'a'..'z' | '\x80'..'\xff'
digit ::= '0'..'9'
IDENTIFIER ::= letter ( ['_'] (letter | digit) )*

Currently any Unicode character with an ordinal value > 127 (non ASCII) is classified as a letter and may thus be part of an identifier but later versions of the language may assign some Unicode characters to belong to the operator characters instead.

The following keywords are reserved and cannot be used as identifiers:

addr and as asm
bind block break
case cast concept const continue converter
defer discard distinct div do
elif else end enum except export
finally for from func
if import in include interface is isnot iterator
macro method mixin mod
nil not notin
object of or out
proc ptr
raise ref return
shl shr static
template try tuple type
when while

Some keywords are unused; they are reserved for future developments of the language.

Identifier equality

Two identifiers are considered equal if the following algorithm returns true:

proc sameIdentifier(a, b: string): bool =
  a[0] == b[0] and
    a.replace("_", "").toLowerAscii == b.replace("_", "").toLowerAscii

That means only the first letters are compared in a case sensitive manner. Other letters are compared case insensitively within the ASCII range and underscores are ignored.

This rather unorthodox way to do identifier comparisons is called partial case insensitivity and has some advantages over the conventional case sensitivity:

It allows programmers to mostly use their own preferred spelling style, be it humpStyle or snake_style, and libraries written by different programmers cannot use incompatible conventions. A Nim-aware editor or IDE can show the identifiers as preferred. Another advantage is that it frees the programmer from remembering the exact spelling of an identifier. The exception with respect to the first letter allows common code like var foo: Foo to be parsed unambiguously.

Historically, Nim was a fully style-insensitive language. This meant that it was not case-sensitive and underscores were ignored and there was not even a distinction between foo and Foo.

String literals

Terminal symbol in the grammar: STR_LIT.

String literals can be delimited by matching double quotes, and can contain the following escape sequences:

Escape sequence Meaning
\p platform specific newline: CRLF on Windows, LF on Unix
\r, \c carriage return
\n, \l line feed (often called newline)
\f form feed
\t tabulator
\v vertical tabulator
\\ backslash
\" quotation mark
\' apostrophe
\ '0'..'9'+ character with decimal value d; all decimal digits directly following are used for the character
\a alert
\b backspace
\e escape [ESC]
\x HH character with hex value HH; exactly two hex digits are allowed

Strings in Nim may contain any 8-bit value, even embedded zeros. However some operations may interpret the first binary zero as a terminator.

Triple quoted string literals

Terminal symbol in the grammar: TRIPLESTR_LIT.

String literals can also be delimited by three double quotes """ ... """. Literals in this form may run for several lines, may contain " and do not interpret any escape sequences. For convenience, when the opening """ is followed by a newline (there may be whitespace between the opening """ and the newline), the newline (and the preceding whitespace) is not included in the string. The ending of the string literal is defined by the pattern """[^"], so this:

""""long string within quotes""""


"long string within quotes"

Raw string literals

Terminal symbol in the grammar: RSTR_LIT.

There are also raw string literals that are preceded with the letter r (or R) and are delimited by matching double quotes (just like ordinary string literals) and do not interpret the escape sequences. This is especially convenient for regular expressions or Windows paths:

var f = openFile(r"C:\texts\text.txt") # a raw string, so ``\t`` is no tab

To produce a single " within a raw string literal, it has to be doubled:




r"""" is not possible with this notation, because the three leading quotes introduce a triple quoted string literal. r""" is the same as """ since triple quoted string literals do not interpret escape sequences either.

Generalized raw string literals


The construct identifier"string literal" (without whitespace between the identifier and the opening quotation mark) is a generalized raw string literal. It is a shortcut for the construct identifier(r"string literal"), so it denotes a procedure call with a raw string literal as its only argument. Generalized raw string literals are especially convenient for embedding mini languages directly into Nim (for example regular expressions).

The construct identifier"""string literal""" exists too. It is a shortcut for identifier("""string literal""").

Character literals

Character literals are enclosed in single quotes '' and can contain the same escape sequences as strings - with one exception: the platform dependent newline (\p) is not allowed as it may be wider than one character (often it is the pair CR/LF for example). Here are the valid escape sequences for character literals:

Escape sequence Meaning
\r, \c carriage return
\n, \l line feed
\f form feed
\t tabulator
\v vertical tabulator
\\ backslash
\" quotation mark
\' apostrophe
\ '0'..'9'+ character with decimal value d; all decimal digits directly following are used for the character
\a alert
\b backspace
\e escape [ESC]
\x HH character with hex value HH; exactly two hex digits are allowed

A character is not an Unicode character but a single byte. The reason for this is efficiency: for the overwhelming majority of use-cases, the resulting programs will still handle UTF-8 properly as UTF-8 was specially designed for this. Another reason is that Nim can thus support array[char, int] or set[char] efficiently as many algorithms rely on this feature. The Rune type is used for Unicode characters, it can represent any Unicode character. Rune is declared in the unicode module.

Numerical constants

Numerical constants are of a single type and have the form:

hexdigit = digit | 'A'..'F' | 'a'..'f'
octdigit = '0'..'7'
bindigit = '0'..'1'
HEX_LIT = '0' ('x' | 'X' ) hexdigit ( ['_'] hexdigit )*
DEC_LIT = digit ( ['_'] digit )*
OCT_LIT = '0' 'o' octdigit ( ['_'] octdigit )*
BIN_LIT = '0' ('b' | 'B' ) bindigit ( ['_'] bindigit )*

        | DEC_LIT
        | OCT_LIT
        | BIN_LIT

INT8_LIT = INT_LIT ['\''] ('i' | 'I') '8'
INT16_LIT = INT_LIT ['\''] ('i' | 'I') '16'
INT32_LIT = INT_LIT ['\''] ('i' | 'I') '32'
INT64_LIT = INT_LIT ['\''] ('i' | 'I') '64'

UINT_LIT = INT_LIT ['\''] ('u' | 'U')
UINT8_LIT = INT_LIT ['\''] ('u' | 'U') '8'
UINT16_LIT = INT_LIT ['\''] ('u' | 'U') '16'
UINT32_LIT = INT_LIT ['\''] ('u' | 'U') '32'
UINT64_LIT = INT_LIT ['\''] ('u' | 'U') '64'

exponent = ('e' | 'E' ) ['+' | '-'] digit ( ['_'] digit )*
FLOAT_LIT = digit (['_'] digit)* (('.' digit (['_'] digit)* [exponent]) |exponent)
FLOAT32_SUFFIX = ('f' | 'F') ['32']
            | (FLOAT_LIT | DEC_LIT | OCT_LIT | BIN_LIT) ['\''] FLOAT32_SUFFIX
FLOAT64_SUFFIX = ( ('f' | 'F') '64' ) | 'd' | 'D'
            | (FLOAT_LIT | DEC_LIT | OCT_LIT | BIN_LIT) ['\''] FLOAT64_SUFFIX

As can be seen in the productions, numerical constants can contain underscores for readability. Integer and floating point literals may be given in decimal (no prefix), binary (prefix 0b), octal (prefix 0o) and hexadecimal (prefix 0x) notation.

There exists a literal for each numerical type that is defined. The suffix starting with an apostrophe (''') is called a type suffix. Literals without a type suffix are of an integer type, unless the literal contains a dot or E|e in which case it is of type float. This integer type is int if the literal is in the range low(i32)..high(i32), otherwise it is int64. For notational convenience the apostrophe of a type suffix is optional if it is not ambiguous (only hexadecimal floating point literals with a type suffix can be ambiguous).

The type suffixes are:

Type Suffix Resulting type of literal
'i8 int8
'i16 int16
'i32 int32
'i64 int64
'u uint
'u8 uint8
'u16 uint16
'u32 uint32
'u64 uint64
'f float32
'd float64
'f32 float32
'f64 float64
'f128 float128

Floating point literals may also be in binary, octal or hexadecimal notation: 0B0_10001110100_0000101001000111101011101111111011000101001101001001'f64 is approximately 1.72826e35 according to the IEEE floating point standard.

Literals are bounds checked so that they fit the datatype. Non base-10 literals are used mainly for flags and bit pattern representations, therefore bounds checking is done on bit width, not value range. If the literal fits in the bit width of the datatype, it is accepted. Hence: 0b10000000'u8 == 0x80'u8 == 128, but, 0b10000000'i8 == 0x80'i8 == -1 instead of causing an overflow error.


Nim allows user defined operators. An operator is any combination of the following characters:

=     +     -     *     /     <     >
@     $     ~     &     %     |
!     ?     ^     .     :     \

These keywords are also operators: and or not xor shl shr div mod in notin is isnot of.

. =, :, :: are not available as general operators; they are used for other notational purposes.

*: is as a special case treated as the two tokens * and : (to support var v*: T).

Other tokens

The following strings denote other tokens:

`   (    )     {    }     [    ]    ,  ;   [.    .]  {.   .}  (.  .)  [:

The slice operator .. takes precedence over other tokens that contain a dot: {..} are the three tokens {, .., } and not the two tokens {., .}.


This section lists Nim's standard syntax. How the parser handles the indentation is already described in the Lexical Analysis section.

Nim allows user-definable operators. Binary operators have 11 different levels of precedence.


Binary operators whose first character is ^ are right-associative, all other binary operators are left-associative.

proc `^/`(x, y: float): float =
  # a right-associative division operator
  result = x / y
echo 12 ^/ 4 ^/ 8 # 24.0 (4 / 8 = 0.5, then 12 / 0.5 = 24.0)
echo 12  / 4  / 8 # 0.375 (12 / 4 = 3.0, then 3 / 8 = 0.375)


Unary operators always bind stronger than any binary operator: $a + b is ($a) + b and not $(a + b).

If an unary operator's first character is @ it is a sigil-like operator which binds stronger than a primarySuffix: @x.abc is parsed as (@x).abc whereas $x.abc is parsed as $(x.abc).

For binary operators that are not keywords the precedence is determined by the following rules:

Operators ending in either ->, ~> or => are called arrow like, and have the lowest precedence of all operators.

If the operator ends with = and its first character is none of <, >, !, =, ~, ?, it is an assignment operator which has the second lowest precedence.

Otherwise precedence is determined by the first character.

Precedence level Operators First character Terminal symbol
10 (highest) $ ^ OP10
9 * / div mod shl shr % * % \ / OP9
8 + - + - ~ | OP8
7 & & OP7
6 .. . OP6
5 == <= < >= > != in notin is isnot not of = < > ! OP5
4 and OP4
3 or xor OP3
2 @ : ? OP2
1 assignment operator (like +=, *=) OP1
0 (lowest) arrow like operator (like ->, =>) OP0

Whether an operator is used a prefix operator is also affected by preceding whitespace (this parsing change was introduced with version 0.13.0):

echo $foo
# is parsed as

Spacing also determines whether (a, b) is parsed as an the argument list of a call or whether it is parsed as a tuple constructor:

echo(1, 2) # pass 1 and 2 to echo
echo (1, 2) # pass the tuple (1, 2) to echo


The grammar's start symbol is module.

# This file is generated by compiler/parser.nim.
module = stmt ^* (';' / IND{=})
comma = ',' COMMENT?
semicolon = ';' COMMENT?
colon = ':' COMMENT?
colcom = ':' COMMENT?
operator =  OP0 | OP1 | OP2 | OP3 | OP4 | OP5 | OP6 | OP7 | OP8 | OP9
         | 'or' | 'xor' | 'and'
         | 'is' | 'isnot' | 'in' | 'notin' | 'of'
         | 'div' | 'mod' | 'shl' | 'shr' | 'not' | 'static' | '..'
prefixOperator = operator
optInd = COMMENT?
optPar = (IND{>} | IND{=})?
simpleExpr = arrowExpr (OP0 optInd arrowExpr)* pragma?
arrowExpr = assignExpr (OP1 optInd assignExpr)*
assignExpr = orExpr (OP2 optInd orExpr)*
orExpr = andExpr (OP3 optInd andExpr)*
andExpr = cmpExpr (OP4 optInd cmpExpr)*
cmpExpr = sliceExpr (OP5 optInd sliceExpr)*
sliceExpr = ampExpr (OP6 optInd ampExpr)*
ampExpr = plusExpr (OP7 optInd plusExpr)*
plusExpr = mulExpr (OP8 optInd mulExpr)*
mulExpr = dollarExpr (OP9 optInd dollarExpr)*
dollarExpr = primary (OP10 optInd primary)*
symbol = '`' (KEYW|IDENT|literal|(operator|'('|')'|'['|']'|'{'|'}'|'=')+)+ '`'
       | IDENT | KEYW
exprColonEqExpr = expr (':'|'=' expr)?
exprList = expr ^+ comma
exprColonEqExprList = exprColonEqExpr (comma exprColonEqExpr)* (comma)?
dotExpr = expr '.' optInd (symbol | '[:' exprList ']')
explicitGenericInstantiation = '[:' exprList ']' ( '(' exprColonEqExpr ')' )?
qualifiedIdent = symbol ('.' optInd symbol)?
setOrTableConstr = '{' ((exprColonEqExpr comma)* | ':' ) '}'
castExpr = 'cast' '[' optInd typeDesc optPar ']' '(' optInd expr optPar ')'
parKeyw = 'discard' | 'include' | 'if' | 'while' | 'case' | 'try'
        | 'finally' | 'except' | 'for' | 'block' | 'const' | 'let'
        | 'when' | 'var' | 'mixin'
par = '(' optInd
          ( &parKeyw complexOrSimpleStmt ^+ ';'
          | ';' complexOrSimpleStmt ^+ ';'
          | pragmaStmt
          | simpleExpr ( ('=' expr (';' complexOrSimpleStmt ^+ ';' )? )
                       | (':' expr (',' exprColonEqExpr     ^+ ',' )? ) ) )
          optPar ')'
literal = | INT_LIT | INT8_LIT | INT16_LIT | INT32_LIT | INT64_LIT
          | UINT_LIT | UINT8_LIT | UINT16_LIT | UINT32_LIT | UINT64_LIT
          | FLOAT_LIT | FLOAT32_LIT | FLOAT64_LIT
          | CHAR_LIT
          | NIL
identOrLiteral = generalizedLit | symbol | literal
               | par | arrayConstr | setOrTableConstr
               | castExpr
tupleConstr = '(' optInd (exprColonEqExpr comma?)* optPar ')'
arrayConstr = '[' optInd (exprColonEqExpr comma?)* optPar ']'
primarySuffix = '(' (exprColonEqExpr comma?)* ')' doBlocks?
      | doBlocks
      | '.' optInd symbol generalizedLit?
      | '[' optInd indexExprList optPar ']'
      | '{' optInd indexExprList optPar '}'
      | &( '`'|IDENT|literal|'cast'|'addr'|'type') expr # command syntax
condExpr = expr colcom expr optInd
        ('elif' expr colcom expr optInd)*
         'else' colcom expr
ifExpr = 'if' condExpr
whenExpr = 'when' condExpr
pragma = '{.' optInd (exprColonExpr comma?)* optPar ('.}' | '}')
identVis = symbol opr?  # postfix position
identVisDot = symbol '.' optInd symbol opr?
identWithPragma = identVis pragma?
identWithPragmaDot = identVisDot pragma?
declColonEquals = identWithPragma (comma identWithPragma)* comma?
                  (':' optInd typeDesc)? ('=' optInd expr)?
identColonEquals = ident (comma ident)* comma?
     (':' optInd typeDesc)? ('=' optInd expr)?)
inlTupleDecl = 'tuple'
    [' optInd  (identColonEquals (comma/semicolon)?)*  optPar ']'
extTupleDecl = 'tuple'
    COMMENT? (IND{>} identColonEquals (IND{=} identColonEquals)*)?
tupleClass = 'tuple'
paramList = '(' declColonEquals ^* (comma/semicolon) ')'
paramListArrow = paramList? ('->' optInd typeDesc)?
paramListColon = paramList? (':' optInd typeDesc)?
doBlock = 'do' paramListArrow pragmas? colcom stmt
procExpr = 'proc' paramListColon pragmas? ('=' COMMENT? stmt)?
distinct = 'distinct' optInd typeDesc
expr = (blockExpr
      | ifExpr
      | whenExpr
      | caseExpr
      | tryExpr)
      / simpleExpr
typeKeyw = 'var' | 'out' | 'ref' | 'ptr' | 'shared' | 'tuple'
         | 'proc' | 'iterator' | 'distinct' | 'object' | 'enum'
primary = typeKeyw typeDescK
        /  prefixOperator* identOrLiteral primarySuffix*
        / 'static' primary
        / 'bind' primary
typeDesc = simpleExpr
typeDefAux = simpleExpr
           | 'concept' typeClass
postExprBlocks = ':' stmt? ( IND{=} doBlock
                           | IND{=} 'of' exprList ':' stmt
                           | IND{=} 'elif' expr ':' stmt
                           | IND{=} 'except' exprList ':' stmt
                           | IND{=} 'else' ':' stmt )*
exprStmt = simpleExpr
         (( '=' optInd expr colonBody? )
         / ( expr ^+ comma
              / macroColon
importStmt = 'import' optInd expr
              ((comma expr)*
              / 'except' optInd (expr ^+ comma))
includeStmt = 'include' optInd expr ^+ comma
fromStmt = 'from' moduleName 'import' optInd expr (comma expr)*
returnStmt = 'return' optInd expr?
raiseStmt = 'raise' optInd expr?
yieldStmt = 'yield' optInd expr?
discardStmt = 'discard' optInd expr?
breakStmt = 'break' optInd expr?
continueStmt = 'break' optInd expr?
condStmt = expr colcom stmt COMMENT?
           (IND{=} 'elif' expr colcom stmt)*
           (IND{=} 'else' colcom stmt)?
ifStmt = 'if' condStmt
whenStmt = 'when' condStmt
whileStmt = 'while' expr colcom stmt
ofBranch = 'of' exprList colcom stmt
ofBranches = ofBranch (IND{=} ofBranch)*
                      (IND{=} 'elif' expr colcom stmt)*
                      (IND{=} 'else' colcom stmt)?
caseStmt = 'case' expr ':'? COMMENT?
            (IND{>} ofBranches DED
            | IND{=} ofBranches)
tryStmt = 'try' colcom stmt &(IND{=}? 'except'|'finally')
           (IND{=}? 'except' exprList colcom stmt)*
           (IND{=}? 'finally' colcom stmt)?
tryExpr = 'try' colcom stmt &(optInd 'except'|'finally')
           (optInd 'except' exprList colcom stmt)*
           (optInd 'finally' colcom stmt)?
exceptBlock = 'except' colcom stmt
forStmt = 'for' (identWithPragma ^+ comma) 'in' expr colcom stmt
blockStmt = 'block' symbol? colcom stmt
blockExpr = 'block' symbol? colcom stmt
staticStmt = 'static' colcom stmt
deferStmt = 'defer' colcom stmt
asmStmt = 'asm' pragma? (STR_LIT | RSTR_LIT | TRIPLE_STR_LIT)
genericParam = symbol (comma symbol)* (colon expr)? ('=' optInd expr)?
genericParamList = '[' optInd
  genericParam ^* (comma/semicolon) optPar ']'
pattern = '{' stmt '}'
indAndComment = (IND{>} COMMENT)? | COMMENT?
routine = optInd identVis pattern? genericParamList?
  paramListColon pragma? ('=' COMMENT? stmt)? indAndComment
commentStmt = COMMENT
section(p) = COMMENT? p / (IND{>} (p / COMMENT)^+IND{=} DED)
constant = identWithPragma (colon typeDesc)? '=' optInd expr indAndComment
enum = 'enum' optInd (symbol optInd ('=' optInd expr COMMENT?)? comma?)+
objectWhen = 'when' expr colcom objectPart COMMENT?
            ('elif' expr colcom objectPart COMMENT?)*
            ('else' colcom objectPart COMMENT?)?
objectBranch = 'of' exprList colcom objectPart
objectBranches = objectBranch (IND{=} objectBranch)*
                      (IND{=} 'elif' expr colcom objectPart)*
                      (IND{=} 'else' colcom objectPart)?
objectCase = 'case' identWithPragma ':' typeDesc ':'? COMMENT?
            (IND{>} objectBranches DED
            | IND{=} objectBranches)
objectPart = IND{>} objectPart^+IND{=} DED
           / objectWhen / objectCase / 'nil' / 'discard' / declColonEquals
object = 'object' pragma? ('of' typeDesc)? COMMENT? objectPart
typeClassParam = ('var' | 'out')? symbol
typeClass = typeClassParam ^* ',' (pragma)? ('of' typeDesc ^* ',')?
              &IND{>} stmt
typeDef = identWithPragmaDot genericParamList? '=' optInd typeDefAux
varTuple = '(' optInd identWithPragma ^+ comma optPar ')' '=' optInd expr
colonBody = colcom stmt doBlocks?
variable = (varTuple / identColonEquals) colonBody? indAndComment
bindStmt = 'bind' optInd qualifiedIdent ^+ comma
mixinStmt = 'mixin' optInd qualifiedIdent ^+ comma
pragmaStmt = pragma (':' COMMENT? stmt)?
simpleStmt = ((returnStmt | raiseStmt | yieldStmt | discardStmt | breakStmt
           | continueStmt | pragmaStmt | importStmt | exportStmt | fromStmt
           | includeStmt | commentStmt) / exprStmt) COMMENT?
complexOrSimpleStmt = (ifStmt | whenStmt | whileStmt
                    | tryStmt | forStmt
                    | blockStmt | staticStmt | deferStmt | asmStmt
                    | 'proc' routine
                    | 'method' routine
                    | 'iterator' routine
                    | 'macro' routine
                    | 'template' routine
                    | 'converter' routine
                    | 'type' section(typeDef)
                    | 'const' section(constant)
                    | ('let' | 'var' | 'using') section(variable)
                    | bindStmt | mixinStmt)
                    / simpleStmt
stmt = (IND{>} complexOrSimpleStmt^+(IND{=} / ';') DED)
     / simpleStmt ^+ ';'

Order of evaluation

Order of evaluation is strictly left-to-right, inside-out as it is typical for most others imperative programming languages:

var s = ""

proc p(arg: int): int =
  s.add $arg
  result = arg

discard p(p(1) + p(2))

doAssert s == "123"

Assignments are not special, the left-hand-side expression is evaluated before the right-hand side:

var v = 0
proc getI(): int =
  result = v
  inc v

var a, b: array[0..2, int]

proc someCopy(a: var int; b: int) = a = b

a[getI()] = getI()

doAssert a == [1, 0, 0]

v = 0
someCopy(b[getI()], getI())

doAssert b == [1, 0, 0]

Rationale: Consistency with overloaded assignment or assignment-like operations, a = b can be read as performSomeCopy(a, b).


All expressions have a type which is known at compile time. Nim is statically typed. One can declare new types, which is in essence defining an identifier that can be used to denote this custom type.

These are the major type classes:

  • ordinal types (consist of integer, bool, character, enumeration (and subranges thereof) types)
  • floating point types
  • string type
  • structured types
  • reference (pointer) type
  • procedural type
  • generic type

Ordinal types

Ordinal types have the following characteristics:

  • Ordinal types are countable and ordered. This property allows the operation of functions as inc, ord, dec on ordinal types to be defined.
  • Ordinal values have a smallest possible value. Trying to count further down than the smallest value gives a checked runtime or static error.
  • Ordinal values have a largest possible value. Trying to count further than the largest value gives a checked runtime or static error.

Integers, bool, characters and enumeration types (and subranges of these types) belong to ordinal types. For reasons of simplicity of implementation the types uint and uint64 are not ordinal types.

Pre-defined integer types

These integer types are pre-defined:

the generic signed integer type; its size is platform dependent and has the same size as a pointer. This type should be used in general. An integer literal that has no type suffix is of this type if it is in the range low(int32)..high(int32) otherwise the literal's type is int64.
additional signed integer types of XX bits use this naming scheme (example: int16 is a 16 bit wide integer). The current implementation supports int8, int16, int32, int64. Literals of these types have the suffix 'iXX.
the generic unsigned integer type; its size is platform dependent and has the same size as a pointer. An integer literal with the type suffix 'u is of this type.
additional unsigned integer types of XX bits use this naming scheme (example: uint16 is a 16 bit wide unsigned integer). The current implementation supports uint8, uint16, uint32, uint64. Literals of these types have the suffix 'uXX. Unsigned operations all wrap around; they cannot lead to over- or underflow errors.

In addition to the usual arithmetic operators for signed and unsigned integers (+ - * etc.) there are also operators that formally work on signed integers but treat their arguments as unsigned: They are mostly provided for backwards compatibility with older versions of the language that lacked unsigned integer types. These unsigned operations for signed integers use the % suffix as convention:

operation meaning
a +% b unsigned integer addition
a -% b unsigned integer subtraction
a *% b unsigned integer multiplication
a /% b unsigned integer division
a %% b unsigned integer modulo operation
a <% b treat a and b as unsigned and compare
a <=% b treat a and b as unsigned and compare
ze(a) extends the bits of a with zeros until it has the width of the int type
toU8(a) treats a as unsigned and converts it to an unsigned integer of 8 bits (but still the int8 type)
toU16(a) treats a as unsigned and converts it to an unsigned integer of 16 bits (but still the int16 type)
toU32(a) treats a as unsigned and converts it to an unsigned integer of 32 bits (but still the int32 type)

Automatic type conversion is performed in expressions where different kinds of integer types are used: the smaller type is converted to the larger.

A narrowing type conversion converts a larger to a smaller type (for example int32 -> int16. A widening type conversion converts a smaller type to a larger type (for example int16 -> int32). In Nim only widening type conversions are implicit:

var myInt16 = 5i16
var myInt: int
myInt16 + 34     # of type ``int16``
myInt16 + myInt  # of type ``int``
myInt16 + 2i32   # of type ``int32``

However, int literals are implicitly convertible to a smaller integer type if the literal's value fits this smaller type and such a conversion is less expensive than other implicit conversions, so myInt16 + 34 produces an int16 result.

For further details, see Convertible relation.

Subrange types

A subrange type is a range of values from an ordinal or floating point type (the base type). To define a subrange type, one must specify it's limiting values: the lowest and highest value of the type:

  Subrange = range[0..5]
  PositiveFloat = range[0.0..Inf]

Subrange is a subrange of an integer which can only hold the values 0 to 5. PositiveFloat defines a subrange of all positive floating point values. NaN does not belong to any subrange of floating point types. Assigning any other value to a variable of type Subrange is a checked runtime error (or static error if it can be statically determined). Assignments from the base type to one of its subrange types (and vice versa) are allowed.

A subrange type has the same size as its base type (int in the Subrange example).

Pre-defined floating point types

The following floating point types are pre-defined:

the generic floating point type; its size used to be platform dependent, but now it is always mapped to float64. This type should be used in general.
an implementation may define additional floating point types of XX bits using this naming scheme (example: float64 is a 64 bit wide float). The current implementation supports float32 and float64. Literals of these types have the suffix 'fXX.

Automatic type conversion in expressions with different kinds of floating point types is performed: See Convertible relation for further details. Arithmetic performed on floating point types follows the IEEE standard. Integer types are not converted to floating point types automatically and vice versa.

The IEEE standard defines five types of floating-point exceptions:

  • Invalid: operations with mathematically invalid operands, for example 0.0/0.0, sqrt(-1.0), and log(-37.8).
  • Division by zero: divisor is zero and dividend is a finite nonzero number, for example 1.0/0.0.
  • Overflow: operation produces a result that exceeds the range of the exponent, for example MAXDOUBLE+0.0000000000001e308.
  • Underflow: operation produces a result that is too small to be represented as a normal number, for example, MINDOUBLE * MINDOUBLE.
  • Inexact: operation produces a result that cannot be represented with infinite precision, for example, 2.0 / 3.0, log(1.1) and 0.1 in input.

The IEEE exceptions are either ignored at runtime or mapped to the Nim exceptions: FloatInvalidOpError, FloatDivByZeroError, FloatOverflowError, FloatUnderflowError, and FloatInexactError. These exceptions inherit from the FloatingPointError base class.

Nim provides the pragmas nanChecks and infChecks to control whether the IEEE exceptions are ignored or trap a Nim exception:

{.nanChecks: on, infChecks: on.}
var a = 1.0
var b = 0.0
echo b / b # raises FloatInvalidOpError
echo a / b # raises FloatOverflowError

In the current implementation FloatDivByZeroError and FloatInexactError are never raised. FloatOverflowError is raised instead of FloatDivByZeroError. There is also a floatChecks pragma that is a short-cut for the combination of nanChecks and infChecks pragmas. floatChecks are turned off as default.

The only operations that are affected by the floatChecks pragma are the +, -, *, / operators for floating point types.

An implementation should always use the maximum precision available to evaluate floating pointer values at compile time; this means expressions like 0.09'f32 + 0.01'f32 == 0.09'f64 + 0.01'f64 are true.

Boolean type

The boolean type is named bool in Nim and can be one of the two pre-defined values true and false. Conditions in while, if, elif, when-statements need to be of type bool.

This condition holds:

ord(false) == 0 and ord(true) == 1

The operators not, and, or, xor, <, <=, >, >=, !=, == are defined for the bool type. The and and or operators perform short-cut evaluation. Example:

while p != nil and p.name != "xyz":
  # p.name is not evaluated if p == nil
  p = p.next

The size of the bool type is one byte.

Character type

The character type is named char in Nim. Its size is one byte. Thus it cannot represent an UTF-8 character, but a part of it. The reason for this is efficiency: for the overwhelming majority of use-cases, the resulting programs will still handle UTF-8 properly as UTF-8 was specially designed for this. Another reason is that Nim can support array[char, int] or set[char] efficiently as many algorithms rely on this feature. The Rune type is used for Unicode characters, it can represent any Unicode character. Rune is declared in the unicode module.

Enumeration types

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

  Direction = enum
    north, east, south, west

Now the following holds:

ord(north) == 0
ord(east) == 1
ord(south) == 2
ord(west) == 3

# Also allowed:
ord(Direction.west) == 3

Thus, north < east < south < west. The comparison operators can be used with enumeration types. Instead of north etc, the enum value can also be qualified with the enum type that it resides in, Direction.north.

For better interfacing to other programming languages, the fields of enum types can be assigned an explicit ordinal value. However, the ordinal values have to be in ascending order. A field whose ordinal value is not explicitly given is assigned the value of the previous field + 1.

An explicit ordered enum can have holes:

  TokenType = enum
    a = 2, b = 4, c = 89 # holes are valid

However, it is then not an ordinal anymore, so it is not possible to use these enums as an index type for arrays. The procedures inc, dec, succ and pred are not available for them either.

The compiler supports the built-in stringify operator $ for enumerations. The stringify's result can be controlled by explicitly giving the string values to use:

  MyEnum = enum
    valueA = (0, "my value A"),
    valueB = "value B",
    valueC = 2,
    valueD = (3, "abc")

As can be seen from the example, it is possible to both specify a field's ordinal value and its string value by using a tuple. It is also possible to only specify one of them.

An enum can be marked with the pure pragma so that it's fields are added to a special module specific hidden scope that is only queried as the last attempt. Only non-ambiguous symbols are added to this scope. But one can always access these via type qualification written as MyEnum.value:

  MyEnum {.pure.} = enum
    valueA, valueB, valueC, valueD, amb
  OtherEnum {.pure.} = enum
    valueX, valueY, valueZ, amb

echo valueA # MyEnum.valueA
echo amb    # Error: Unclear whether it's MyEnum.amb or OtherEnum.amb
echo MyEnum.amb # OK.

String type

All string literals are of the type string. A string in Nim is very similar to a sequence of characters. However, strings in Nim are both zero-terminated and have a length field. One can retrieve the length with the builtin len procedure; the length never counts the terminating zero.

The terminating zero cannot be accessed unless the string is converted to the cstring type first. The terminating zero assures that this conversion can be done in O(1) and without any allocations.

The assignment operator for strings always copies the string. The & operator concatenates strings.

Most native Nim types support conversion to strings with the special $ proc. When calling the echo proc, for example, the built-in stringify operation for the parameter is called:

echo 3 # calls `$` for `int`

Whenever a user creates a specialized object, implementation of this procedure provides for string representation.

  Person = object
    name: string
    age: int

proc `$`(p: Person): string = # `$` always returns a string
  result = p.name & " is " &
          $p.age & # we *need* the `$` in front of p.age which
                   # is natively an integer to convert it to
                   # a string
          " years old."

While $p.name can also be used, the $ operation on a string does nothing. Note that we cannot rely on automatic conversion from an int to a string like we can for the echo proc.

Strings are compared by their lexicographical order. All comparison operators are available. Strings can be indexed like arrays (lower bound is 0). Unlike arrays, they can be used in case statements:

case paramStr(i)
of "-v": incl(options, optVerbose)
of "-h", "-?": incl(options, optHelp)
else: write(stdout, "invalid command line option!\n")

Per convention, all strings are UTF-8 strings, but this is not enforced. For example, when reading strings from binary files, they are merely a sequence of bytes. The index operation s[i] means the i-th char of s, not the i-th unichar. The iterator runes from the unicode module can be used for iteration over all Unicode characters.

cstring type

The cstring type meaning compatible string is the native representation of a string for the compilation backend. For the C backend the cstring type represents a pointer to a zero-terminated char array compatible to the type char* in Ansi C. Its primary purpose lies in easy interfacing with C. The index operation s[i] means the i-th char of s; however no bounds checking for cstring is performed making the index operation unsafe.

A Nim string is implicitly convertible to cstring for convenience. If a Nim string is passed to a C-style variadic proc, it is implicitly converted to cstring too:

proc printf(formatstr: cstring) {.importc: "printf", varargs,
                                  header: "<stdio.h>".}

printf("This works %s", "as expected")

Even though the conversion is implicit, it is not safe: The garbage collector does not consider a cstring to be a root and may collect the underlying memory. However in practice this almost never happens as the GC considers stack roots conservatively. One can use the builtin procs GC_ref and GC_unref to keep the string data alive for the rare cases where it does not work.

A $ proc is defined for cstrings that returns a string. Thus to get a nim string from a cstring:

var str: string = "Hello!"
var cstr: cstring = str
var newstr: string = $cstr

Structured types

A variable of a structured type can hold multiple values at the same time. Structured types can be nested to unlimited levels. Arrays, sequences, tuples, objects and sets belong to the structured types.

Array and sequence types

Arrays are a homogeneous type, meaning that each element in the array has the same type. Arrays always have a fixed length which is specified at compile time (except for open arrays). They can be indexed by any ordinal type. A parameter A may be an open array, in which case it is indexed by integers from 0 to len(A)-1. An array expression may be constructed by the array constructor []. The element type of this array expression is inferred from the type of the first element. All other elements need to be implicitly convertable to this type.

Sequences are similar to arrays but of dynamic length which may change during runtime (like strings). Sequences are implemented as growable arrays, allocating pieces of memory as items are added. A sequence S is always indexed by integers from 0 to len(S)-1 and its bounds are checked. Sequences can be constructed by the array constructor [] in conjunction with the array to sequence operator @. Another way to allocate space for a sequence is to call the built-in newSeq procedure.

A sequence may be passed to a parameter that is of type open array.


  IntArray = array[0..5, int] # an array that is indexed with 0..5
  IntSeq = seq[int] # a sequence of integers
  x: IntArray
  y: IntSeq
x = [1, 2, 3, 4, 5, 6]  # [] is the array constructor
y = @[1, 2, 3, 4, 5, 6] # the @ turns the array into a sequence

let z = [1.0, 2, 3, 4] # the type of z is array[0..3, float]

The lower bound of an array or sequence may be received by the built-in proc low(), the higher bound by high(). The length may be received by len(). low() for a sequence or an open array always returns 0, as this is the first valid index. One can append elements to a sequence with the add() proc or the & operator, and remove (and get) the last element of a sequence with the pop() proc.

The notation x[i] can be used to access the i-th element of x.

Arrays are always bounds checked (at compile-time or at runtime). These checks can be disabled via pragmas or invoking the compiler with the --boundChecks:off command line switch.

Open arrays

Often fixed size arrays turn out to be too inflexible; procedures should be able to deal with arrays of different sizes. The openarray type allows this; it can only be used for parameters. Openarrays are always indexed with an int starting at position 0. The len, low and high operations are available for open arrays too. Any array with a compatible base type can be passed to an openarray parameter, the index type does not matter. In addition to arrays sequences can also be passed to an open array parameter.

The openarray type cannot be nested: multidimensional openarrays are not supported because this is seldom needed and cannot be done efficiently.

proc testOpenArray(x: openArray[int]) = echo repr(x)

testOpenArray([1,2,3])  # array[]
testOpenArray(@[1,2,3]) # seq[]


A varargs parameter is an openarray parameter that additionally allows to pass a variable number of arguments to a procedure. The compiler converts the list of arguments to an array implicitly:

proc myWriteln(f: File, a: varargs[string]) =
  for s in items(a):
    write(f, s)
  write(f, "\n")

myWriteln(stdout, "abc", "def", "xyz")
# is transformed to:
myWriteln(stdout, ["abc", "def", "xyz"])

This transformation is only done if the varargs parameter is the last parameter in the procedure header. It is also possible to perform type conversions in this context:

proc myWriteln(f: File, a: varargs[string, `$`]) =
  for s in items(a):
    write(f, s)
  write(f, "\n")

myWriteln(stdout, 123, "abc", 4.0)
# is transformed to:
myWriteln(stdout, [$123, $"def", $4.0])

In this example $ is applied to any argument that is passed to the parameter a. (Note that $ applied to strings is a nop.)

Note that an explicit array constructor passed to a varargs parameter is not wrapped in another implicit array construction:

proc takeV[T](a: varargs[T]) = discard

takeV([123, 2, 1]) # takeV's T is "int", not "array of int"

varargs[typed] is treated specially: It matches a variable list of arguments of arbitrary type but always constructs an implicit array. This is required so that the builtin echo proc does what is expected:

proc echo*(x: varargs[typed, `$`]) {...}

echo @[1, 2, 3]
# prints "@[1, 2, 3]" and not "123"

Tuples and object types

A variable of a tuple or object type is a heterogeneous storage container. A tuple or object defines various named fields of a type. A tuple also defines an order of the fields. Tuples are meant for heterogeneous storage types with no overhead and few abstraction possibilities. The constructor () can be used to construct tuples. The order of the fields in the constructor must match the order of the tuple's definition. Different tuple-types are equivalent if they specify the same fields of the same type in the same order. The names of the fields also have to be identical.

The assignment operator for tuples copies each component. The default assignment operator for objects copies each component. Overloading of the assignment operator is described in type-bound-operations-operator.

  Person = tuple[name: string, age: int] # type representing a person:
                                         # a person consists of a name
                                         # and an age
  person: Person
person = (name: "Peter", age: 30)
# the same, but less readable:
person = ("Peter", 30)

A tuple with one unnamed field can be constructed with the parentheses and a trailing comma:

proc echoUnaryTuple(a: (int,)) =
  echo a[0]

echoUnaryTuple (1,)

In fact, a trailing comma is allowed for every tuple construction.

The implementation aligns the fields for best access performance. The alignment is compatible with the way the C compiler does it.

For consistency with object declarations, tuples in a type section can also be defined with indentation instead of []:

  Person = tuple   # type representing a person
    name: string   # a person consists of a name
    age: natural   # and an age

Objects provide many features that tuples do not. Object provide inheritance and information hiding. Objects have access to their type at runtime, so that the of operator can be used to determine the object's type. The of operator is similar to the instanceof operator in Java.

  Person = object of RootObj
    name*: string   # the * means that `name` is accessible from other modules
    age: int        # no * means that the field is hidden
  Student = ref object of Person # a student is a person
    id: int                      # with an id field

  student: Student
  person: Person
assert(student of Student) # is true
assert(student of Person) # also true

Object fields that should be visible from outside the defining module, have to be marked by *. In contrast to tuples, different object types are never equivalent. Objects that have no ancestor are implicitly final and thus have no hidden type field. One can use the inheritable pragma to introduce new object roots apart from system.RootObj.

Object construction

Objects can also be created with an object construction expression that has the syntax T(fieldA: valueA, fieldB: valueB, ...) where T is an object type or a ref object type:

var student = Student(name: "Anton", age: 5, id: 3)

Note that, unlike tuples, objects require the field names along with their values. For a ref object type system.new is invoked implicitly.

Object variants

Often an object hierarchy is overkill in certain situations where simple variant types are needed.

An example:

# This is an example how an abstract syntax tree could be modelled in Nim
  NodeKind = enum  # the different node types
    nkInt,          # a leaf with an integer value
    nkFloat,        # a leaf with a float value
    nkString,       # a leaf with a string value
    nkAdd,          # an addition
    nkSub,          # a subtraction
    nkIf            # an if statement
  Node = ref NodeObj
  NodeObj = object
    case kind: NodeKind  # the ``kind`` field is the discriminator
    of nkInt: intVal: int
    of nkFloat: floatVal: float
    of nkString: strVal: string
    of nkAdd, nkSub:
      leftOp, rightOp: Node
    of nkIf:
      condition, thenPart, elsePart: Node

# create a new case object:
var n = Node(kind: nkIf, condition: nil)
# accessing n.thenPart is valid because the ``nkIf`` branch is active:
n.thenPart = Node(kind: nkFloat, floatVal: 2.0)

# the following statement raises an `FieldError` exception, because
# n.kind's value does not fit and the ``nkString`` branch is not active:
n.strVal = ""

# invalid: would change the active object branch:
n.kind = nkInt

var x = Node(kind: nkAdd, leftOp: Node(kind: nkInt, intVal: 4),
                          rightOp: Node(kind: nkInt, intVal: 2))
# valid: does not change the active object branch:
x.kind = nkSub

As can been seen from the example, an advantage to an object hierarchy is that no casting between different object types is needed. Yet, access to invalid object fields raises an exception.

The syntax of case in an object declaration follows closely the syntax of the case statement: The branches in a case section may be indented too.

In the example the kind field is called the discriminator: For safety its address cannot be taken and assignments to it are restricted: The new value must not lead to a change of the active object branch. For an object branch switch system.reset has to be used. Also, when the fields of a particular branch are specified during object construction, the correct value for the discriminator must be supplied at compile-time.

Package level objects

Every Nim module resides in a (nimble) package. An object type can be attached to the package it resides in. If that is done, the type can be referenced from other modules as an incomplete object type. This features allows to break up recursive type dependencies accross module boundaries. Incomplete object types are always passed byref and can only be used in pointer like contexts (var/ref/ptr IncompleteObject) in general since the compiler does not yet know the size of the object. To complete an incomplete object the package pragma has to be used. package implies byref.

As long as a type T is incomplete sizeof(T) or "runtime type information" for T is not available.


# module A (in an arbitrary package)
  Pack.SomeObject = object ## declare as incomplete object of package 'Pack'
  Triple = object
    a, b, c: ref SomeObject ## pointers to incomplete objects are allowed

## Incomplete objects can be used as parameters:
proc myproc(x: SomeObject) = discard
# module B (in package "Pack")
  SomeObject* {.package.} = object ## Use 'package' to complete the object
    s, t: string
    x, y: int

Set type

The set type models the mathematical notion of a set. The set's basetype can only be an ordinal type of a certain size, namely:
  • int8-int16
  • uint8/byte-uint16
  • char
  • enum

or equivalent. The reason is that sets are implemented as high performance bit vectors. Attempting to declare a set with a larger type will result in an error:

var s: set[int64] # Error: set is too large

Sets can be constructed via the set constructor: {} is the empty set. The empty set is type compatible with any concrete set type. The constructor can also be used to include elements (and ranges of elements):

  CharSet = set[char]
  x: CharSet
x = {'a'..'z', '0'..'9'} # This constructs a set that contains the
                         # letters from 'a' to 'z' and the digits
                         # from '0' to '9'

These operations are supported by sets:

operation meaning
A + B union of two sets
A * B intersection of two sets
A - B difference of two sets (A without B's elements)
A == B set equality
A <= B subset relation (A is subset of B or equal to B)
A < B strong subset relation (A is a real subset of B)
e in A set membership (A contains element e)
e notin A A does not contain element e
contains(A, e) A contains element e
card(A) the cardinality of A (number of elements in A)
incl(A, elem) same as A = A + {elem}
excl(A, elem) same as A = A - {elem}

Sets are often used to define a type for the flags of a procedure. This is a much cleaner (and type safe) solution than just defining integer constants that should be or'ed together.

Reference and pointer types

References (similar to pointers in other programming languages) are a way to introduce many-to-one relationships. This means different references can point to and modify the same location in memory (also called aliasing).

Nim distinguishes between traced and untraced references. Untraced references are also called pointers. Traced references point to objects of a garbage collected heap, untraced references point to manually allocated objects or to objects somewhere else in memory. Thus untraced references are unsafe. However for certain low-level operations (accessing the hardware) untraced references are unavoidable.

Traced references are declared with the ref keyword, untraced references are declared with the ptr keyword. In general, a ptr T is implicitly convertible to the pointer type.

An empty subscript [] notation can be used to derefer a reference, the addr procedure returns the address of an item. An address is always an untraced reference. Thus the usage of addr is an unsafe feature.

The . (access a tuple/object field operator) and [] (array/string/sequence index operator) operators perform implicit dereferencing operations for reference types:

  Node = ref NodeObj
  NodeObj = object
    le, ri: Node
    data: int

  n: Node
n.data = 9
# no need to write n[].data; in fact n[].data is highly discouraged!

Automatic dereferencing is also performed for the first argument of a routine call. But currently this feature has to be only enabled via {.experimental: "implicitDeref".}:

{.experimental: "implicitDeref".}

proc depth(x: NodeObj): int = ...

  n: Node
echo n.depth
# no need to write n[].depth either

In order to simplify structural type checking, recursive tuples are not valid:

# invalid recursion
type MyTuple = tuple[a: ref MyTuple]

Likewise T = ref T is an invalid type.

As a syntactical extension object types can be anonymous if declared in a type section via the ref object or ptr object notations. This feature is useful if an object should only gain reference semantics:

  Node = ref object
    le, ri: Node
    data: int

To allocate a new traced object, the built-in procedure new has to be used. To deal with untraced memory, the procedures alloc, dealloc and realloc can be used. The documentation of the system module contains further information.

If a reference points to nothing, it has the value nil.

Special care has to be taken if an untraced object contains traced objects like traced references, strings or sequences: in order to free everything properly, the built-in procedure GCunref has to be called before freeing the untraced memory manually:

  Data = tuple[x, y: int, s: string]

# allocate memory for Data on the heap:
var d = cast[ptr Data](alloc0(sizeof(Data)))

# create a new string on the garbage collected heap:
d.s = "abc"

# tell the GC that the string is not needed anymore:

# free the memory:

Without the GCunref call the memory allocated for the d.s string would never be freed. The example also demonstrates two important features for low level programming: the sizeof proc returns the size of a type or value in bytes. The cast operator can circumvent the type system: the compiler is forced to treat the result of the alloc0 call (which returns an untyped pointer) as if it would have the type ptr Data. Casting should only be done if it is unavoidable: it breaks type safety and bugs can lead to mysterious crashes.

Note: The example only works because the memory is initialized to zero (alloc0 instead of alloc does this): d.s is thus initialized to binary zero which the string assignment can handle. One needs to know low level details like this when mixing garbage collected data with unmanaged memory.

Not nil annotation

All types for that nil is a valid value can be annotated to exclude nil as a valid value with the not nil annotation:

  PObject = ref TObj not nil
  TProc = (proc (x, y: int)) not nil

proc p(x: PObject) =
  echo "not nil"

# compiler catches this:

# and also this:
var x: PObject

The compiler ensures that every code path initializes variables which contain non nilable pointers. The details of this analysis are still to be specified here.

Procedural type

A procedural type is internally a pointer to a procedure. nil is an allowed value for variables of a procedural type. Nim uses procedural types to achieve functional programming techniques.


proc printItem(x: int) = ...

proc forEach(c: proc (x: int) {.cdecl.}) =

forEach(printItem)  # this will NOT compile because calling conventions differ
  OnMouseMove = proc (x, y: int) {.closure.}

proc onMouseMove(mouseX, mouseY: int) =
  # has default calling convention
  echo "x: ", mouseX, " y: ", mouseY

proc setOnMouseMove(mouseMoveEvent: OnMouseMove) = discard

# ok, 'onMouseMove' has the default calling convention, which is compatible
# to 'closure':

A subtle issue with procedural types is that the calling convention of the procedure influences the type compatibility: procedural types are only compatible if they have the same calling convention. As a special extension, a procedure of the calling convention nimcall can be passed to a parameter that expects a proc of the calling convention closure.

Nim supports these calling conventions:

is the default convention used for a Nim proc. It is the same as fastcall, but only for C compilers that support fastcall.
is the default calling convention for a procedural type that lacks any pragma annotations. It indicates that the procedure has a hidden implicit parameter (an environment). Proc vars that have the calling convention closure take up two machine words: One for the proc pointer and another one for the pointer to implicitly passed environment.
This the stdcall convention as specified by Microsoft. The generated C procedure is declared with the __stdcall keyword.
The cdecl convention means that a procedure shall use the same convention as the C compiler. Under windows the generated C procedure is declared with the __cdecl keyword.
This is the safecall convention as specified by Microsoft. The generated C procedure is declared with the __safecall keyword. The word safe refers to the fact that all hardware registers shall be pushed to the hardware stack.
The inline convention means the the caller should not call the procedure, but inline its code directly. Note that Nim does not inline, but leaves this to the C compiler; it generates __inline procedures. This is only a hint for the compiler: it may completely ignore it and it may inline procedures that are not marked as inline.
Fastcall means different things to different C compilers. One gets whatever the C __fastcall means.
The syscall convention is the same as __syscall in C. It is used for interrupts.
The generated C code will not have any explicit calling convention and thus use the C compiler's default calling convention. This is needed because Nim's default calling convention for procedures is fastcall to improve speed.

Most calling conventions exist only for the Windows 32-bit platform.

The default calling convention is nimcall, unless it is an inner proc (a proc inside of a proc). For an inner proc an analysis is performed whether it accesses its environment. If it does so, it has the calling convention closure, otherwise it has the calling convention nimcall.

Distinct type

A distinct type is new type derived from a base type that is incompatible with its base type. In particular, it is an essential property of a distinct type that it does not imply a subtype relation between it and its base type. Explicit type conversions from a distinct type to its base type and vice versa are allowed. See also distinctBase to get the reverse operation.

Modelling currencies

A distinct type can be used to model different physical units with a numerical base type, for example. The following example models currencies.

Different currencies should not be mixed in monetary calculations. Distinct types are a perfect tool to model different currencies:

  Dollar = distinct int
  Euro = distinct int

  d: Dollar
  e: Euro

echo d + 12
# Error: cannot add a number with no unit and a ``Dollar``

Unfortunately, d + 12.Dollar is not allowed either, because + is defined for int (among others), not for Dollar. So a + for dollars needs to be defined:

proc `+` (x, y: Dollar): Dollar =
  result = Dollar(int(x) + int(y))

It does not make sense to multiply a dollar with a dollar, but with a number without unit; and the same holds for division:

proc `*` (x: Dollar, y: int): Dollar =
  result = Dollar(int(x) * y)

proc `*` (x: int, y: Dollar): Dollar =
  result = Dollar(x * int(y))

proc `div` ...

This quickly gets tedious. The implementations are trivial and the compiler should not generate all this code only to optimize it away later - after all + for dollars should produce the same binary code as + for ints. The pragma borrow has been designed to solve this problem; in principle it generates the above trivial implementations:

proc `*` (x: Dollar, y: int): Dollar {.borrow.}
proc `*` (x: int, y: Dollar): Dollar {.borrow.}
proc `div` (x: Dollar, y: int): Dollar {.borrow.}

The borrow pragma makes the compiler use the same implementation as the proc that deals with the distinct type's base type, so no code is generated.

But it seems all this boilerplate code needs to be repeated for the Euro currency. This can be solved with templates.

template additive(typ: type) =
  proc `+` *(x, y: typ): typ {.borrow.}
  proc `-` *(x, y: typ): typ {.borrow.}
  # unary operators:
  proc `+` *(x: typ): typ {.borrow.}
  proc `-` *(x: typ): typ {.borrow.}

template multiplicative(typ, base: type) =
  proc `*` *(x: typ, y: base): typ {.borrow.}
  proc `*` *(x: base, y: typ): typ {.borrow.}
  proc `div` *(x: typ, y: base): typ {.borrow.}
  proc `mod` *(x: typ, y: base): typ {.borrow.}

template comparable(typ: type) =
  proc `<` * (x, y: typ): bool {.borrow.}
  proc `<=` * (x, y: typ): bool {.borrow.}
  proc `==` * (x, y: typ): bool {.borrow.}

template defineCurrency(typ, base: untyped) =
    typ* = distinct base
  multiplicative(typ, base)

defineCurrency(Dollar, int)
defineCurrency(Euro, int)

The borrow pragma can also be used to annotate the distinct type to allow certain builtin operations to be lifted:

  Foo = object
    a, b: int
    s: string
  Bar {.borrow: `.`.} = distinct Foo

var bb: ref Bar
new bb
# field access now valid
bb.a = 90
bb.s = "abc"

Currently only the dot accessor can be borrowed in this way.

Avoiding SQL injection attacks

An SQL statement that is passed from Nim to an SQL database might be modelled as a string. However, using string templates and filling in the values is vulnerable to the famous SQL injection attack:

import strutils

proc query(db: DbHandle, statement: string) = ...

  username: string

db.query("SELECT FROM users WHERE name = '$1'" % username)
# Horrible security hole, but the compiler does not mind!

This can be avoided by distinguishing strings that contain SQL from strings that don't. Distinct types provide a means to introduce a new string type SQL that is incompatible with string:

  SQL = distinct string

proc query(db: DbHandle, statement: SQL) = ...

  username: string

db.query("SELECT FROM users WHERE name = '$1'" % username)
# Error at compile time: `query` expects an SQL string!

It is an essential property of abstract types that they do not imply a subtype relation between the abstract type and its base type. Explicit type conversions from string to SQL are allowed:

import strutils, sequtils

proc properQuote(s: string): SQL =
  # quotes a string properly for an SQL statement
  return SQL(s)

proc `%` (frmt: SQL, values: openarray[string]): SQL =
  # quote each argument:
  let v = values.mapIt(SQL, properQuote(it))
  # we need a temporary type for the type conversion :-(
  type StrSeq = seq[string]
  # call strutils.`%`:
  result = SQL(string(frmt) % StrSeq(v))

db.query("SELECT FROM users WHERE name = '$1'".SQL % [username])

Now we have compile-time checking against SQL injection attacks. Since "".SQL is transformed to SQL("") no new syntax is needed for nice looking SQL string literals. The hypothetical SQL type actually exists in the library as the TSqlQuery type of modules like db_sqlite.

Void type

The void type denotes the absence of any type. Parameters of type void are treated as non-existent, void as a return type means that the procedure does not return a value:

proc nothing(x, y: void): void =
  echo "ha"

nothing() # writes "ha" to stdout

The void type is particularly useful for generic code:

proc callProc[T](p: proc (x: T), x: T) =
  when T is void:

proc intProc(x: int) = discard
proc emptyProc() = discard

callProc[int](intProc, 12)

However, a void type cannot be inferred in generic code:

# Error: type mismatch: got (proc ())
# but expected one of:
# callProc(p: proc (T), x: T)

The void type is only valid for parameters and return types; other symbols cannot have the type void.

Auto type

The auto type can only be used for return types and parameters. For return types it causes the compiler to infer the type from the routine body:

proc returnsInt(): auto = 1984

For parameters it currently creates implicitly generic routines:

proc foo(a, b: auto) = discard

Is the same as:

proc foo[T1, T2](a: T1, b: T2) = discard

However later versions of the language might change this to mean "infer the parameters' types from the body". Then the above foo would be rejected as the parameters' types can not be inferred from an empty discard statement.

Type relations

The following section defines several relations on types that are needed to describe the type checking done by the compiler.

Type equality

Nim uses structural type equivalence for most types. Only for objects, enumerations and distinct types name equivalence is used. The following algorithm, in pseudo-code, determines type equality:

proc typeEqualsAux(a, b: PType,
                   s: var HashSet[(PType, PType)]): bool =
  if (a,b) in s: return true
  incl(s, (a,b))
  if a.kind == b.kind:
    case a.kind
    of int, intXX, float, floatXX, char, string, cstring, pointer,
        bool, nil, void:
      # leaf type: kinds identical; nothing more to check
      result = true
    of ref, ptr, var, set, seq, openarray:
      result = typeEqualsAux(a.baseType, b.baseType, s)
    of range:
      result = typeEqualsAux(a.baseType, b.baseType, s) and
        (a.rangeA == b.rangeA) and (a.rangeB == b.rangeB)
    of array:
      result = typeEqualsAux(a.baseType, b.baseType, s) and
               typeEqualsAux(a.indexType, b.indexType, s)
    of tuple:
      if a.tupleLen == b.tupleLen:
        for i in 0..a.tupleLen-1:
          if not typeEqualsAux(a[i], b[i], s): return false
        result = true
    of object, enum, distinct:
      result = a == b
    of proc:
      result = typeEqualsAux(a.parameterTuple, b.parameterTuple, s) and
               typeEqualsAux(a.resultType, b.resultType, s) and
               a.callingConvention == b.callingConvention

proc typeEquals(a, b: PType): bool =
  var s: HashSet[(PType, PType)] = {}
  result = typeEqualsAux(a, b, s)

Since types are graphs which can have cycles, the above algorithm needs an auxiliary set s to detect this case.

Type equality modulo type distinction

The following algorithm (in pseudo-code) determines whether two types are equal with no respect to distinct types. For brevity the cycle check with an auxiliary set s is omitted:

proc typeEqualsOrDistinct(a, b: PType): bool =
  if a.kind == b.kind:
    case a.kind
    of int, intXX, float, floatXX, char, string, cstring, pointer,
        bool, nil, void:
      # leaf type: kinds identical; nothing more to check
      result = true
    of ref, ptr, var, set, seq, openarray:
      result = typeEqualsOrDistinct(a.baseType, b.baseType)
    of range:
      result = typeEqualsOrDistinct(a.baseType, b.baseType) and
        (a.rangeA == b.rangeA) and (a.rangeB == b.rangeB)
    of array:
      result = typeEqualsOrDistinct(a.baseType, b.baseType) and
               typeEqualsOrDistinct(a.indexType, b.indexType)
    of tuple:
      if a.tupleLen == b.tupleLen:
        for i in 0..a.tupleLen-1:
          if not typeEqualsOrDistinct(a[i], b[i]): return false
        result = true
    of distinct:
      result = typeEqualsOrDistinct(a.baseType, b.baseType)
    of object, enum:
      result = a == b
    of proc:
      result = typeEqualsOrDistinct(a.parameterTuple, b.parameterTuple) and
               typeEqualsOrDistinct(a.resultType, b.resultType) and
               a.callingConvention == b.callingConvention
  elif a.kind == distinct:
    result = typeEqualsOrDistinct(a.baseType, b)
  elif b.kind == distinct:
    result = typeEqualsOrDistinct(a, b.baseType)

Subtype relation

If object a inherits from b, a is a subtype of b. This subtype relation is extended to the types var, ref, ptr:

proc isSubtype(a, b: PType): bool =
  if a.kind == b.kind:
    case a.kind
    of object:
      var aa = a.baseType
      while aa != nil and aa != b: aa = aa.baseType
      result = aa == b
    of var, ref, ptr:
      result = isSubtype(a.baseType, b.baseType)


Covariance in Nim can be introduced only though pointer-like types such as ptr and ref. Sequence, Array and OpenArray types, instantiated with pointer-like types will be considered covariant if and only if they are also immutable. The introduction of a var modifier or additional ptr or ref indirections would result in invariant treatment of these types.

proc types are currently always invariant, but future versions of Nim may relax this rule.

User-defined generic types may also be covariant with respect to some of their parameters. By default, all generic params are considered invariant, but you may choose the apply the prefix modifier in to a parameter to make it contravariant or out to make it covariant:

  AnnotatedPtr[out T] =
    metadata: MyTypeInfo
    p: ref T
  RingBuffer[out T] =
    startPos: int
    data: seq[T]
  Action {.importcpp: "std::function<void ('0)>".} [in T] = object

When the designated generic parameter is used to instantiate a pointer-like type as in the case of AnnotatedPtr above, the resulting generic type will also have pointer-like covariance:

  GuiWidget = object of RootObj
  Button = object of GuiWidget
  ComboBox = object of GuiWidget

  widgetPtr: AnnotatedPtr[GuiWidget]
  buttonPtr: AnnotatedPtr[Button]


proc drawWidget[T](x: AnnotatedPtr[GuiWidget]) = ...

# you can call procs expecting base types by supplying a derived type

# and you can convert more-specific pointer types to more general ones
widgetPtr = buttonPtr

Just like with regular pointers, covariance will be enabled only for immutable values:

proc makeComboBox[T](x: var AnnotatedPtr[GuiWidget]) =
  x.p = new(ComboBox)

makeComboBox(buttonPtr) # Error, AnnotatedPtr[Button] cannot be modified
                        # to point to a ComboBox

On the other hand, in the RingBuffer example above, the designated generic param is used to instantiate the non-pointer seq type, which means that the resulting generic type will have covariance that mimics an array or sequence (i.e. it will be covariant only when instantiated with ptr and ref types):

  Base = object of RootObj
  Derived = object of Base

proc consumeBaseValues(b: RingBuffer[Base]) = ...

var derivedValues: RingBuffer[Derived]

consumeBaseValues(derivedValues) # Error, Base and Derived values may differ
                                 # in size

proc consumeBasePointers(b: RingBuffer[ptr Base]) = ...

var derivedPointers: RingBuffer[ptr Derived]

consumeBaseValues(derivedPointers) # This is legal

Please note that Nim will treat the user-defined pointer-like types as proper alternatives to the built-in pointer types. That is, types such as seq[AnnotatedPtr[T]] or RingBuffer[AnnotatedPtr[T]] will also be considered covariant and you can create new pointer-like types by instantiating other user-defined pointer-like types.

The contravariant parameters introduced with the in modifier are currently useful only when interfacing with imported types having such semantics.

Convertible relation

A type a is implicitly convertible to type b iff the following algorithm returns true:

# XXX range types?
proc isImplicitlyConvertible(a, b: PType): bool =
  if isSubtype(a, b) or isCovariant(a, b):
    return true
  case a.kind
  of int:     result = b in {int8, int16, int32, int64, uint, uint8, uint16,
                             uint32, uint64, float, float32, float64}
  of int8:    result = b in {int16, int32, int64, int}
  of int16:   result = b in {int32, int64, int}
  of int32:   result = b in {int64, int}
  of uint:    result = b in {uint32, uint64}
  of uint8:   result = b in {uint16, uint32, uint64}
  of uint16:  result = b in {uint32, uint64}
  of uint32:  result = b in {uint64}
  of float:   result = b in {float32, float64}
  of float32: result = b in {float64, float}
  of float64: result = b in {float32, float}
  of seq:
    result = b == openArray and typeEquals(a.baseType, b.baseType)
  of array:
    result = b == openArray and typeEquals(a.baseType, b.baseType)
    if a.baseType == char and a.indexType.rangeA == 0:
      result = b = cstring
  of cstring, ptr:
    result = b == pointer
  of string:
    result = b == cstring

A type a is explicitly convertible to type b iff the following algorithm returns true:

proc isIntegralType(t: PType): bool =
  result = isOrdinal(t) or t.kind in {float, float32, float64}

proc isExplicitlyConvertible(a, b: PType): bool =
  result = false
  if isImplicitlyConvertible(a, b): return true
  if typeEqualsOrDistinct(a, b): return true
  if isIntegralType(a) and isIntegralType(b): return true
  if isSubtype(a, b) or isSubtype(b, a): return true

The convertible relation can be relaxed by a user-defined type converter.

converter toInt(x: char): int = result = ord(x)

  x: int
  chr: char = 'a'

# implicit conversion magic happens here
x = chr
echo x # => 97
# you can use the explicit form too
x = chr.toInt
echo x # => 97

The type conversion T(a) is an L-value if a is an L-value and typeEqualsOrDistinct(T, type(a)) holds.

Assignment compatibility

An expression b can be assigned to an expression a iff a is an l-value and isImplicitlyConvertible(b.typ, a.typ) holds.

Overloading resolution

In a call p(args) the routine p that matches best is selected. If multiple routines match equally well, the ambiguity is reported at compiletime.

Every arg in args needs to match. There are multiple different categories how an argument can match. Let f be the formal parameter's type and a the type of the argument.

  1. Exact match: a and f are of the same type.
  2. Literal match: a is an integer literal of value v and f is a signed or unsigned integer type and v is in f's range. Or: a is a floating point literal of value v and f is a floating point type and v is in f's range.
  3. Generic match: f is a generic type and a matches, for instance a is int and f is a generic (constrained) parameter type (like in [T] or [T: int|char].
  4. Subrange or subtype match: a is a range[T] and T matches f exactly. Or: a is a subtype of f.
  5. Integral conversion match: a is convertible to f and f and a is some integer or floating point type.
  6. Conversion match: a is convertible to f, possibly via a user defined converter.

These matching categories have a priority: An exact match is better than a literal match and that is better than a generic match etc. In the following count(p, m) counts the number of matches of the matching category m for the routine p.

A routine p matches better than a routine q if the following algorithm returns true:

for each matching category m in ["exact match", "literal match",
                                "generic match", "subtype match",
                                "integral match", "conversion match"]:
  if count(p, m) > count(q, m): return true
  elif count(p, m) == count(q, m):
    discard "continue with next category m"
    return false
return "ambiguous"

Some examples:

proc takesInt(x: int) = echo "int"
proc takesInt[T](x: T) = echo "T"
proc takesInt(x: int16) = echo "int16"

takesInt(4) # "int"
var x: int32
takesInt(x) # "T"
var y: int16
takesInt(y) # "int16"
var z: range[0..4] = 0
takesInt(z) # "T"

If this algorithm returns "ambiguous" further disambiguation is performed: If the argument a matches both the parameter type f of p and g of q via a subtyping relation, the inheritance depth is taken into account:

  A = object of RootObj
  B = object of A
  C = object of B

proc p(obj: A) =
  echo "A"

proc p(obj: B) =
  echo "B"

var c = C()
# not ambiguous, calls 'B', not 'A' since B is a subtype of A
# but not vice versa:

proc pp(obj: A, obj2: B) = echo "A B"
proc pp(obj: B, obj2: A) = echo "B A"

# but this is ambiguous:
pp(c, c)

Likewise for generic matches the most specialized generic type (that still matches) is preferred:

proc gen[T](x: ref ref T) = echo "ref ref T"
proc gen[T](x: ref T) = echo "ref T"
proc gen[T](x: T) = echo "T"

var ri: ref int
gen(ri) # "ref T"

Overloading based on 'var T'

If the formal parameter f is of type var T in addition to the ordinary type checking, the argument is checked to be an l-value. var T matches better than just T then.

proc sayHi(x: int): string =
  # matches a non-var int
  result = $x
proc sayHi(x: var int): string =
  # matches a var int
  result = $(x + 10)

proc sayHello(x: int) =
  var m = x # a mutable version of x
  echo sayHi(x) # matches the non-var version of sayHi
  echo sayHi(m) # matches the var version of sayHi

sayHello(3) # 3
            # 13

Automatic dereferencing

If the experimental mode is active and no other match is found, the first argument a is dereferenced automatically if it's a pointer type and overloading resolution is tried with a[] instead.

Automatic self insertions

Note: The .this pragma is deprecated and should not be used anymore.

Starting with version 0.14 of the language, Nim supports field as a shortcut for self.field comparable to the this keyword in Java or C++. This feature has to be explicitly enabled via a {.this: self.} statement pragma. This pragma is active for the rest of the module:

  Parent = object of RootObj
    parentField: int
  Child = object of Parent
    childField: int

{.this: self.}
proc sumFields(self: Child): int =
  result = parentField + childField
  # is rewritten to:
  # result = self.parentField + self.childField

Instead of self any other identifier can be used too, but {.this: self.} will become the default directive for the whole language eventually.

In addition to fields, routine applications are also rewritten, but only if no other interpretation of the call is possible:

proc test(self: Child) =
  echo childField, " ", sumFields()
  # is rewritten to:
  echo self.childField, " ", sumFields(self)
  # but NOT rewritten to:
  echo self, self.childField, " ", sumFields(self)

Lazy type resolution for untyped

Note: An unresolved expression is an expression for which no symbol lookups and no type checking have been performed.

Since templates and macros that are not declared as immediate participate in overloading resolution it's essential to have a way to pass unresolved expressions to a template or macro. This is what the meta-type untyped accomplishes:

template rem(x: untyped) = discard

rem unresolvedExpression(undeclaredIdentifier)

A parameter of type untyped always matches any argument (as long as there is any argument passed to it).

But one has to watch out because other overloads might trigger the argument's resolution:

template rem(x: untyped) = discard
proc rem[T](x: T) = discard

# undeclared identifier: 'unresolvedExpression'
rem unresolvedExpression(undeclaredIdentifier)

untyped and varargs[untyped] are the only metatype that are lazy in this sense, the other metatypes typed and type are not lazy.

Varargs matching

See Varargs.

Statements and expressions

Nim uses the common statement/expression paradigm: Statements do not produce a value in contrast to expressions. However, some expressions are statements.

Statements are separated into simple statements and complex statements. Simple statements are statements that cannot contain other statements like assignments, calls or the return statement; complex statements can contain other statements. To avoid the dangling else problem, complex statements always have to be indented. The details can be found in the grammar.

Statement list expression

Statements can also occur in an expression context that looks like (stmt1; stmt2; ...; ex). This is called an statement list expression or (;). The type of (stmt1; stmt2; ...; ex) is the type of ex. All the other statements must be of type void. (One can use discard to produce a void type.) (;) does not introduce a new scope.

Discard statement


proc p(x, y: int): int =
  result = x + y

discard p(3, 4) # discard the return value of `p`

The discard statement evaluates its expression for side-effects and throws the expression's resulting value away.

Ignoring the return value of a procedure without using a discard statement is a static error.

The return value can be ignored implicitly if the called proc/iterator has been declared with the discardable pragma:

proc p(x, y: int): int {.discardable.} =
  result = x + y

p(3, 4) # now valid

An empty discard statement is often used as a null statement:

proc classify(s: string) =
  case s[0]
  of SymChars, '_': echo "an identifier"
  of '0'..'9': echo "a number"
  else: discard

Void context

In a list of statements every expression except the last one needs to have the type void. In addition to this rule an assignment to the builtin result symbol also triggers a mandatory void context for the subsequent expressions:

proc invalid*(): string =
  result = "foo"
  "invalid"  # Error: value of type 'string' has to be discarded
proc valid*(): string =
  let x = 317

Var statement

Var statements declare new local and global variables and initialize them. A comma separated list of variables can be used to specify variables of the same type:

  a: int = 0
  x, y, z: int

If an initializer is given the type can be omitted: the variable is then of the same type as the initializing expression. Variables are always initialized with a default value if there is no initializing expression. The default value depends on the type and is always a zero in binary.

Type default value
any integer type 0
any float 0.0
char '\0'
bool false
ref or pointer type nil
procedural type nil
sequence @[]
string ""
tuple[x: A, y: B, ...] (default(A), default(B), ...) (analogous for objects)
array[0..., T] [default(T), ...]
range[T] default(T); this may be out of the valid range
T = enum cast[T](0); this may be an invalid value

The implicit initialization can be avoided for optimization reasons with the noinit pragma:

  a {.noInit.}: array[0..1023, char]

If a proc is annotated with the noinit pragma this refers to its implicit result variable:

proc returnUndefinedValue: int {.noinit.} = discard

The implicit initialization can be also prevented by the requiresInit type pragma. The compiler requires an explicit initialization for the object and all of its fields. However it does a control flow analysis to prove the variable has been initialized and does not rely on syntactic properties:

  MyObject = object {.requiresInit.}

proc p() =
  # the following is valid:
  var x: MyObject
  if someCondition():
    x = a()
    x = a()
  # use x

let statement

A let statement declares new local and global single assignment variables and binds a value to them. The syntax is the same as that of the var statement, except that the keyword var is replaced by the keyword let. Let variables are not l-values and can thus not be passed to var parameters nor can their address be taken. They cannot be assigned new values.

For let variables the same pragmas are available as for ordinary variables.

Tuple unpacking

In a var or let statement tuple unpacking can be performed. The special identifier _ can be used to ignore some parts of the tuple:

proc returnsTuple(): (int, int, int) = (4, 2, 3)

let (x, _, z) = returnsTuple()

Const section

Constants are symbols which are bound to a value. The constant's value cannot change. The compiler must be able to evaluate the expression in a constant declaration at compile time.

Nim contains a sophisticated compile-time evaluator, so procedures which have no side-effect can be used in constant expressions too:

import strutils
  constEval = contains("abc", 'b') # computed at compile time!

The rules for compile-time computability are:

  1. Literals are compile-time computable.
  2. Type conversions are compile-time computable.
  3. Procedure calls of the form p(X) are compile-time computable if p is a proc without side-effects (see the noSideEffect pragma for details) and if X is a (possibly empty) list of compile-time computable arguments.

Constants cannot be of type ptr, ref or var, nor can they contain such a type.

Static statement/expression

A static statement/expression can be used to enforce compile time evaluation explicitly. Enforced compile time evaluation can even evaluate code that has side effects:

  echo "echo at compile time"

It's a static error if the compiler cannot perform the evaluation at compile time.

The current implementation poses some restrictions for compile time evaluation: Code which contains cast or makes use of the foreign function interface cannot be evaluated at compile time. Later versions of Nim will support the FFI at compile time.

If statement


var name = readLine(stdin)

if name == "Andreas":
  echo "What a nice name!"
elif name == "":
  echo "Don't you have a name?"
  echo "Boring name..."

The if statement is a simple way to make a branch in the control flow: The expression after the keyword if is evaluated, if it is true the corresponding statements after the : are executed. Otherwise the expression after the elif is evaluated (if there is an elif branch), if it is true the corresponding statements after the : are executed. This goes on until the last elif. If all conditions fail, the else part is executed. If there is no else part, execution continues with the next statement.

In if statements new scopes begin immediately after the if/elif/else keywords and ends after the corresponding then block. For visualization purposes the scopes have been enclosed in {| |} in the following example:

if {| (let m = input =~ re"(\w+)=\w+"; m.isMatch):
  echo "key ", m[0], " value ", m[1]  |}
elif {| (let m = input =~ re""; m.isMatch):
  echo "new m in this scope"  |}
else: {|
  echo "m not declared here"  |}

Case statement


case readline(stdin)
of "delete-everything", "restart-computer":
  echo "permission denied"
of "go-for-a-walk":     echo "please yourself"
else:                   echo "unknown command"

# indentation of the branches is also allowed; and so is an optional colon
# after the selecting expression:
case readline(stdin):
  of "delete-everything", "restart-computer":
    echo "permission denied"
  of "go-for-a-walk":     echo "please yourself"
  else:                   echo "unknown command"

The case statement is similar to the if statement, but it represents a multi-branch selection. The expression after the keyword case is evaluated and if its value is in a slicelist the corresponding statements (after the of keyword) are executed. If the value is not in any given slicelist the else part is executed. If there is no else part and not all possible values that expr can hold occur in a slicelist, a static error occurs. This holds only for expressions of ordinal types. "All possible values" of expr are determined by expr's type. To suppress the static error an else part with an empty discard statement should be used.

For non ordinal types it is not possible to list every possible value and so these always require an else part.

As case statements perform compile-time exhaustiveness checks, the value in every of branch must be known at compile time. This fact is also exploited to generate more performant code.

As a special semantic extension, an expression in an of branch of a case statement may evaluate to a set or array constructor; the set or array is then expanded into a list of its elements:

  SymChars: set[char] = {'a'..'z', 'A'..'Z', '\x80'..'\xFF'}

proc classify(s: string) =
  case s[0]
  of SymChars, '_': echo "an identifier"
  of '0'..'9': echo "a number"
  else: echo "other"

# is equivalent to:
proc classify(s: string) =
  case s[0]
  of 'a'..'z', 'A'..'Z', '\x80'..'\xFF', '_': echo "an identifier"
  of '0'..'9': echo "a number"
  else: echo "other"

When statement


when sizeof(int) == 2:
  echo "running on a 16 bit system!"
elif sizeof(int) == 4:
  echo "running on a 32 bit system!"
elif sizeof(int) == 8:
  echo "running on a 64 bit system!"
  echo "cannot happen!"

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

  • Each condition (expr) has to be a constant expression (of type bool).
  • The statements do not open a new scope.
  • The statements that belong to the expression that evaluated to true are translated by the compiler, the other statements are not checked for semantics! However, each condition is checked for semantics.

The when statement enables conditional compilation techniques. As a special syntactic extension, the when construct is also available within object definitions.

When nimvm statement

nimvm is a special symbol, that may be used as expression of when nimvm statement to differentiate execution path between runtime and compile time.


proc someProcThatMayRunInCompileTime(): bool =
  when nimvm:
    # This code runs in compile time
    result = true
    # This code runs in runtime
    result = false
const ctValue = someProcThatMayRunInCompileTime()
let rtValue = someProcThatMayRunInCompileTime()
assert(ctValue == true)
assert(rtValue == false)

when nimvm statement must meet the following requirements:

  • Its expression must always be nimvm. More complex expressions are not allowed.
  • It must not contain elif branches.
  • It must contain else branch.
  • Code in branches must not affect semantics of the code that follows the when nimvm statement. E.g. it must not define symbols that are used in the following code.

Return statement


return 40+2

The return statement ends the execution of the current procedure. It is only allowed in procedures. If there is an expr, this is syntactic sugar for:

result = expr
return result

return without an expression is a short notation for return result if the proc has a return type. The result variable is always the return value of the procedure. It is automatically declared by the compiler. As all variables, result is initialized to (binary) zero:

proc returnZero(): int =
  # implicitly returns 0

Yield statement


yield (1, 2, 3)

The yield statement is used instead of the return statement in iterators. It is only valid in iterators. Execution is returned to the body of the for loop that called the iterator. Yield does not end the iteration process, but execution is passed back to the iterator if the next iteration starts. See the section about iterators (Iterators and the for statement) for further information.

Block statement


var found = false
block myblock:
  for i in 0..3:
    for j in 0..3:
      if a[j][i] == 7:
        found = true
        break myblock # leave the block, in this case both for-loops
echo found

The block statement is a means to group statements to a (named) block. Inside the block, the break statement is allowed to leave the block immediately. A break statement can contain a name of a surrounding block to specify which block is to leave.

Break statement



The break statement is used to leave a block immediately. If symbol is given, it is the name of the enclosing block that is to leave. If it is absent, the innermost block is left.

While statement


echo "Please tell me your password:"
var pw = readLine(stdin)
while pw != "12345":
  echo "Wrong password! Next try:"
  pw = readLine(stdin)

The while statement is executed until the expr evaluates to false. Endless loops are no error. while statements open an implicit block, so that they can be left with a break statement.

Continue statement

A continue statement leads to the immediate next iteration of the surrounding loop construct. It is only allowed within a loop. A continue statement is syntactic sugar for a nested block:

while expr1:

Is equivalent to:

while expr1:
  block myBlockName:
    break myBlockName

Assembler statement

The direct embedding of assembler code into Nim code is supported by the unsafe asm statement. Identifiers in the assembler code that refer to Nim identifiers shall be enclosed in a special character which can be specified in the statement's pragmas. The default special character is '`':

{.push stackTrace:off.}
proc addInt(a, b: int): int =
  # a in eax, and b in edx
  asm """
      mov eax, `a`
      add eax, `b`
      jno theEnd
      call `raiseOverflow`

If the GNU assembler is used, quotes and newlines are inserted automatically:

proc addInt(a, b: int): int =
  asm """
    addl %%ecx, %%eax
    jno 1
    call `raiseOverflow`
    :"a"(`a`), "c"(`b`)

Instead of:

proc addInt(a, b: int): int =
  asm """
    "addl %%ecx, %%eax\n"
    "jno 1\n"
    "call `raiseOverflow`\n"
    "1: \n"
    :"a"(`a`), "c"(`b`)

Using statement

The using statement provides syntactic convenience in modules where the same parameter names and types are used over and over. Instead of:

proc foo(c: Context; n: Node) = ...
proc bar(c: Context; n: Node, counter: int) = ...
proc baz(c: Context; n: Node) = ...

One can tell the compiler about the convention that a parameter of name c should default to type Context, n should default to Node etc.:

  c: Context
  n: Node
  counter: int

proc foo(c, n) = ...
proc bar(c, n, counter) = ...
proc baz(c, n) = ...

proc mixedMode(c, n; x, y: int) =
  # 'c' is inferred to be of the type 'Context'
  # 'n' is inferred to be of the type 'Node'
  # But 'x' and 'y' are of type 'int'.

The using section uses the same indentation based grouping syntax as a var or let section.

Note that using is not applied for template since untyped template parameters default to the type system.untyped.

Mixing parameters that should use the using declaration with parameters that are explicitly typed is possible and requires a semicolon between them.

If expression

An if expression is almost like an if statement, but it is an expression. Example:

var y = if x > 8: 9 else: 10

An if expression always results in a value, so the else part is required. Elif parts are also allowed.

When expression

Just like an if expression, but corresponding to the when statement.

Case expression

The case expression is again very similar to the case statement:

var favoriteFood = case animal
  of "dog": "bones"
  of "cat": "mice"
  elif animal.endsWith"whale": "plankton"
    echo "I'm not sure what to serve, but everybody loves ice cream"
    "ice cream"

As seen in the above example, the case expression can also introduce side effects. When multiple statements are given for a branch, Nim will use the last expression as the result value, much like in an expr template.

Table constructor

A table constructor is syntactic sugar for an array constructor:

{"key1": "value1", "key2", "key3": "value2"}

# is the same as:
[("key1", "value1"), ("key2", "value2"), ("key3", "value2")]

The empty table can be written {:} (in contrast to the empty set which is {}) which is thus another way to write as the empty array constructor []. This slightly unusual way of supporting tables has lots of advantages:

  • The order of the (key,value)-pairs is preserved, thus it is easy to support ordered dicts with for example {key: val}.newOrderedTable.
  • A table literal can be put into a const section and the compiler can easily put it into the executable's data section just like it can for arrays and the generated data section requires a minimal amount of memory.
  • Every table implementation is treated equal syntactically.
  • Apart from the minimal syntactic sugar the language core does not need to know about tables.

Type conversions

Syntactically a type conversion is like a procedure call, but a type name replaces the procedure name. A type conversion is always safe in the sense that a failure to convert a type to another results in an exception (if it cannot be determined statically).

Ordinary procs are often preferred over type conversions in Nim: For instance, $ is the toString operator by convention and toFloat and toInt can be used to convert from floating point to integer or vice versa.

Type casts



Type casts are a crude mechanism to interpret the bit pattern of an expression as if it would be of another type. Type casts are only needed for low-level programming and are inherently unsafe.

The addr operator

The addr operator returns the address of an l-value. If the type of the location is T, the addr operator result is of the type ptr T. An address is always an untraced reference. Taking the address of an object that resides on the stack is unsafe, as the pointer may live longer than the object on the stack and can thus reference a non-existing object. One can get the address of variables, but one can't use it on variables declared through let statements:

let t1 = "Hello"
  t2 = t1
  t3 : pointer = addr(t2)
echo repr(addr(t2))
# --> ref 0x7fff6b71b670 --> 0x10bb81050"Hello"
echo cast[ptr string](t3)[]
# --> Hello
# The following line doesn't compile:
echo repr(addr(t1))
# Error: expression has no address

The unsafeAddr operator

For easier interoperability with other compiled languages such as C, retrieving the address of a let variable, a parameter or a for loop variable, the unsafeAddr operation can be used:

let myArray = [1, 2, 3]
foreignProcThatTakesAnAddr(unsafeAddr myArray)


What most programming languages call methods or functions are called procedures in Nim. A procedure declaration consists of an identifier, zero or more formal parameters, a return value type and a block of code. Formal parameters are declared as a list of identifiers separated by either comma or semicolon. A parameter is given a type by : typename. The type applies to all parameters immediately before it, until either the beginning of the parameter list, a semicolon separator or an already typed parameter, is reached. The semicolon can be used to make separation of types and subsequent identifiers more distinct.

# Using only commas
proc foo(a, b: int, c, d: bool): int

# Using semicolon for visual distinction
proc foo(a, b: int; c, d: bool): int

# Will fail: a is untyped since ';' stops type propagation.
proc foo(a; b: int; c, d: bool): int

A parameter may be declared with a default value which is used if the caller does not provide a value for the argument.

# b is optional with 47 as its default value
proc foo(a: int, b: int = 47): int

Parameters can be declared mutable and so allow the proc to modify those arguments, by using the type modifier var.

# "returning" a value to the caller through the 2nd argument
# Notice that the function uses no actual return value at all (ie void)
proc foo(inp: int, outp: var int) =
  outp = inp + 47

If the proc declaration has no body, it is a forward declaration. If the proc returns a value, the procedure body can access an implicitly declared variable named result that represents the return value. Procs can be overloaded. The overloading resolution algorithm determines which proc is the best match for the arguments. Example:

proc toLower(c: char): char = # toLower for characters
  if c in {'A'..'Z'}:
    result = chr(ord(c) + (ord('a') - ord('A')))
    result = c

proc toLower(s: string): string = # toLower for strings
  result = newString(len(s))
  for i in 0..len(s) - 1:
    result[i] = toLower(s[i]) # calls toLower for characters; no recursion!

Calling a procedure can be done in many different ways:

proc callme(x, y: int, s: string = "", c: char, b: bool = false) = ...

# call with positional arguments      # parameter bindings:
callme(0, 1, "abc", '\t', true)       # (x=0, y=1, s="abc", c='\t', b=true)
# call with named and positional arguments:
callme(y=1, x=0, "abd", '\t')         # (x=0, y=1, s="abd", c='\t', b=false)
# call with named arguments (order is not relevant):
callme(c='\t', y=1, x=0)              # (x=0, y=1, s="", c='\t', b=false)
# call as a command statement: no () needed:
callme 0, 1, "abc", '\t'              # (x=0, y=1, s="abc", c='\t', b=false)

A procedure may call itself recursively.

Operators are procedures with a special operator symbol as identifier:

proc `$` (x: int): string =
  # converts an integer to a string; this is a prefix operator.
  result = intToStr(x)

Operators with one parameter are prefix operators, operators with two parameters are infix operators. (However, the parser distinguishes these from the operator's position within an expression.) There is no way to declare postfix operators: all postfix operators are built-in and handled by the grammar explicitly.

Any operator can be called like an ordinary proc with the 'opr' notation. (Thus an operator can have more than two parameters):

proc `*+` (a, b, c: int): int =
  # Multiply and add
  result = a * b + c

assert `*+`(3, 4, 6) == `+`(`*`(a, b), c)

Export marker

If a declared symbol is marked with an asterisk it is exported from the current module:

proc exportedEcho*(s: string) = echo s
proc `*`*(a: string; b: int): string =
  result = newStringOfCap(a.len * b)
  for i in 1..b: result.add a

var exportedVar*: int
const exportedConst* = 78
  ExportedType* = object
    exportedField*: int

Method call syntax

For object oriented programming, the syntax obj.method(args) can be used instead of method(obj, args). The parentheses can be omitted if there are no remaining arguments: obj.len (instead of len(obj)).

This method call syntax is not restricted to objects, it can be used to supply any type of first argument for procedures:

echo "abc".len # is the same as echo len "abc"
echo "abc".toUpper()
echo {'a', 'b', 'c'}.card
stdout.writeLine("Hallo") # the same as writeLine(stdout, "Hallo")

Another way to look at the method call syntax is that it provides the missing postfix notation.

The method call syntax conflicts with explicit generic instantiations: p[T](x) cannot be written as x.p[T] because x.p[T] is always parsed as (x.p)[T].

See also: Limitations of the method call syntax.

The [: ] notation has been designed to mitigate this issue: x.p[:T] is rewritten by the parser to p[T](x), x.p[:T](y) is rewritten to p[T](x, y). Note that [: ] has no AST representation, the rewrite is performed directly in the parsing step.


Nim has no need for get-properties: Ordinary get-procedures that are called with the method call syntax achieve the same. But setting a value is different; for this a special setter syntax is needed:

# Module asocket
  Socket* = ref object of RootObj
    host: int # cannot be accessed from the outside of the module

proc `host=`*(s: var Socket, value: int) {.inline.} =
  ## setter of hostAddr.
  ## This accesses the 'host' field and is not a recursive call to
  ## ``host=`` because the builtin dot access is preferred if it is
  ## avaliable:
  s.host = value

proc host*(s: Socket): int {.inline.} =
  ## getter of hostAddr
  ## This accesses the 'host' field and is not a recursive call to
  ## ``host`` because the builtin dot access is preferred if it is
  ## avaliable:
# module B
import asocket
var s: Socket
new s
s.host = 34  # same as `host=`(s, 34)

Command invocation syntax

Routines can be invoked without the () if the call is syntactically a statement. This command invocation syntax also works for expressions, but then only a single argument may follow. This restriction means echo f 1, f 2 is parsed as echo(f(1), f(2)) and not as echo(f(1, f(2))). The method call syntax may be used to provide one more argument in this case:

proc optarg(x: int, y: int = 0): int = x + y
proc singlearg(x: int): int = 20*x

echo optarg 1, " ", singlearg 2  # prints "1 40"

let fail = optarg 1, optarg 8   # Wrong. Too many arguments for a command call
let x = optarg(1, optarg 8)  # traditional procedure call with 2 arguments
let y = 1.optarg optarg 8    # same thing as above, w/o the parenthesis
assert x == y

The command invocation syntax also can't have complex expressions as arguments. For example: (anonymous procs), if, case or try. The (do notation) is limited, but usable for a single proc (see the example in the corresponding section). Function calls with no arguments still needs () to distinguish between a call and the function itself as a first class value.


Procedures can appear at the top level in a module as well as inside other scopes, in which case they are called nested procs. A nested proc can access local variables from its enclosing scope and if it does so it becomes a closure. Any captured variables are stored in a hidden additional argument to the closure (its environment) and they are accessed by reference by both the closure and its enclosing scope (i.e. any modifications made to them are visible in both places). The closure environment may be allocated on the heap or on the stack if the compiler determines that this would be safe.

Creating closures in loops

Since closures capture local variables by reference it is often not wanted behavior inside loop bodies. See closureScope for details on how to change this behavior.

Anonymous Procs

Procs can also be treated as expressions, in which case it's allowed to omit the proc's name.

var cities = @["Frankfurt", "Tokyo", "New York", "Kyiv"]

cities.sort(proc (x,y: string): int =
    cmp(x.len, y.len))

Procs as expressions can appear both as nested procs and inside top level executable code.


The func keyword introduces a shortcut for a noSideEffect proc.

func binarySearch[T](a: openArray[T]; elem: T): int

Is short for:

proc binarySearch[T](a: openArray[T]; elem: T): int {.noSideEffect.}

Do notation

As a special more convenient notation, proc expressions involved in procedure calls can use the do keyword:

sort(cities) do (x,y: string) -> int:
  cmp(x.len, y.len)

# Less parenthesis using the method plus command syntax:
cities = cities.map do (x:string) -> string:
  "City of " & x

# In macros, the do notation is often used for quasi-quoting
macroResults.add quote do:
  if not `ex`:
    echo `info`, ": Check failed: ", `expString`

do is written after the parentheses enclosing the regular proc params. The proc expression represented by the do block is appended to them. In calls using the command syntax, the do block will bind to the immediately preceeding expression, transforming it in a call.

do with parentheses is an anonymous proc; however a do without parentheses is just a block of code. The do notation can be used to pass multiple blocks to a macro:

macro performWithUndo(task, undo: untyped) = ...

performWithUndo do:
  # multiple-line block of code
  # to perform the task
  # code to undo it

Nonoverloadable builtins

The following builtin procs cannot be overloaded for reasons of implementation simplicity (they require specialized semantic checking):

declared, defined, definedInScope, compiles, sizeOf,
is, shallowCopy, getAst, astToStr, spawn, procCall

Thus they act more like keywords than like ordinary identifiers; unlike a keyword however, a redefinition may shadow the definition in the system module. From this list the following should not be written in dot notation x.f since x cannot be type checked before it gets passed to f:

declared, defined, definedInScope, compiles, getAst, astToStr

Var parameters

The type of a parameter may be prefixed with the var keyword:

proc divmod(a, b: int; res, remainder: var int) =
  res = a div b
  remainder = a mod b

  x, y: int

divmod(8, 5, x, y) # modifies x and y
assert x == 1
assert y == 3

In the example, res and remainder are var parameters. Var parameters can be modified by the procedure and the changes are visible to the caller. The argument passed to a var parameter has to be an l-value. Var parameters are implemented as hidden pointers. The above example is equivalent to:

proc divmod(a, b: int; res, remainder: ptr int) =
  res[] = a div b
  remainder[] = a mod b

  x, y: int
divmod(8, 5, addr(x), addr(y))
assert x == 1
assert y == 3

In the examples, var parameters or pointers are used to provide two return values. This can be done in a cleaner way by returning a tuple:

proc divmod(a, b: int): tuple[res, remainder: int] =
  (a div b, a mod b)

var t = divmod(8, 5)

assert t.res == 1
assert t.remainder == 3

One can use tuple unpacking to access the tuple's fields:

var (x, y) = divmod(8, 5) # tuple unpacking
assert x == 1
assert y == 3

Note: var parameters are never necessary for efficient parameter passing. Since non-var parameters cannot be modified the compiler is always free to pass arguments by reference if it considers it can speed up execution.

Var return type

A proc, converter or iterator may return a var type which means that the returned value is an l-value and can be modified by the caller:

var g = 0

proc writeAccessToG(): var int =
  result = g

writeAccessToG() = 6
assert g == 6

It is a compile time error if the implicitly introduced pointer could be used to access a location beyond its lifetime:

proc writeAccessToG(): var int =
  var g = 0
  result = g # Error!

For iterators, a component of a tuple return type can have a var type too:

iterator mpairs(a: var seq[string]): tuple[key: int, val: var string] =
  for i in 0..a.high:
    yield (i, a[i])

In the standard library every name of a routine that returns a var type starts with the prefix m per convention.

Memory safety for returning by var T is ensured by a simple borrowing rule: If result does not refer to a location pointing to the heap (that is in result = X the X involves a ptr or ref access) then it has to be deviated by the routine's first parameter:
proc forward[T](x: var T): var T =
  result = x # ok, deviated from the first parameter.

proc p(param: var int): var int =
  var x: int
  # we know 'forward' provides a view into the location deviated by
  # its first argument 'x'.
  result = forward(x) # Error: location is derived from ``x``
                      # which is not p's first parameter and lives
                      # on the stack.

In other words, the lifetime of what result points to is attached to the lifetime of the first parameter and that is enough knowledge to verify memory safety at the callsite.

Future directions

Later versions of Nim can be more precise about the borrowing rule with a syntax like:

proc foo(other: Y; container: var X): var T from container

Here var T from container explicitly exposes that the location is deviated from the second parameter (called 'container' in this case). The syntax var T from p specifies a type varTy[T, 2] which is incompatible with varTy[T, 1].

Overloading of the subscript operator

The [] subscript operator for arrays/openarrays/sequences can be overloaded.


Procedures always use static dispatch. Multi-methods use dynamic dispatch. For dynamic dispatch to work on an object it should be a reference type as well.

  Expression = ref object of RootObj ## abstract base class for an expression
  Literal = ref object of Expression
    x: int
  PlusExpr = ref object of Expression
    a, b: Expression

method eval(e: Expression): int {.base.} =
  # override this base method
  quit "to override!"

method eval(e: Literal): int = return e.x

method eval(e: PlusExpr): int =
  # watch out: relies on dynamic binding
  result = eval(e.a) + eval(e.b)

proc newLit(x: int): Literal =
  result.x = x

proc newPlus(a, b: Expression): PlusExpr =
  result.a = a
  result.b = b

echo eval(newPlus(newPlus(newLit(1), newLit(2)), newLit(4)))

In the example the constructors newLit and newPlus are procs because they should use static binding, but eval is a method because it requires dynamic binding.

As can be seen in the example, base methods have to be annotated with the base pragma. The base pragma also acts as a reminder for the programmer that a base method m is used as the foundation to determine all the effects that a call to m might cause.

In a multi-method all parameters that have an object type are used for the dispatching:

  Thing = ref object of RootObj
  Unit = ref object of Thing
    x: int

method collide(a, b: Thing) {.base, inline.} =
  quit "to override!"

method collide(a: Thing, b: Unit) {.inline.} =
  echo "1"

method collide(a: Unit, b: Thing) {.inline.} =
  echo "2"

var a, b: Unit
new a
new b
collide(a, b) # output: 2

Invocation of a multi-method cannot be ambiguous: collide 2 is preferred over collide 1 because the resolution works from left to right. In the example Unit, Thing is preferred over Thing, Unit.

Note: Compile time evaluation is not (yet) supported for methods.

Inhibit dynamic method resolution via procCall

Dynamic method resolution can be inhibited via the builtin system.procCall. This is somewhat comparable to the super keyword that traditional OOP languages offer.

  Thing = ref object of RootObj
  Unit = ref object of Thing
    x: int

method m(a: Thing) {.base.} =
  echo "base"

method m(a: Unit) =
  # Call the base method:
  procCall m(Thing(a))
  echo "1"

Iterators and the for statement

The for statement is an abstract mechanism to iterate over the elements of a container. It relies on an iterator to do so. Like while statements, for statements open an implicit block, so that they can be left with a break statement.

The for loop declares iteration variables - their scope reaches until the end of the loop body. The iteration variables' types are inferred by the return type of the iterator.

An iterator is similar to a procedure, except that it can be called in the context of a for loop. Iterators provide a way to specify the iteration over an abstract type. A key role in the execution of a for loop plays the yield statement in the called iterator. Whenever a yield statement is reached the data is bound to the for loop variables and control continues in the body of the for loop. The iterator's local variables and execution state are automatically saved between calls. Example:

# this definition exists in the system module
iterator items*(a: string): char {.inline.} =
  var i = 0
  while i < len(a):
    yield a[i]

for ch in items("hello world"): # `ch` is an iteration variable
  echo ch

The compiler generates code as if the programmer would have written this:

var i = 0
while i < len(a):
  var ch = a[i]
  echo ch

If the iterator yields a tuple, there can be as many iteration variables as there are components in the tuple. The i'th iteration variable's type is the type of the i'th component. In other words, implicit tuple unpacking in a for loop context is supported.

Implict items/pairs invocations

If the for loop expression e does not denote an iterator and the for loop has exactly 1 variable, the for loop expression is rewritten to items(e); ie. an items iterator is implicitly invoked:

for x in [1,2,3]: echo x

If the for loop has exactly 2 variables, a pairs iterator is implicitly invoked.

Symbol lookup of the identifiers items/pairs is performed after the rewriting step, so that all overloads of items/pairs are taken into account.

First class iterators

There are 2 kinds of iterators in Nim: inline and closure iterators. An inline iterator is an iterator that's always inlined by the compiler leading to zero overhead for the abstraction, but may result in a heavy increase in code size. Inline iterators are second class citizens; They can be passed as parameters only to other inlining code facilities like templates, macros and other inline iterators.

In contrast to that, a closure iterator can be passed around more freely:

iterator count0(): int {.closure.} =
  yield 0

iterator count2(): int {.closure.} =
  var x = 1
  yield x
  inc x
  yield x

proc invoke(iter: iterator(): int {.closure.}) =
  for x in iter(): echo x


Closure iterators have other restrictions than inline iterators:

  1. yield in a closure iterator can not occur in a try statement.
  2. For now, a closure iterator cannot be evaluated at compile time.
  3. return is allowed in a closure iterator (but rarely useful) and ends iteration.
  4. Neither inline nor closure iterators can be recursive.
  5. Closure iterators are not supported by the js backend.

Iterators that are neither marked {.closure.} nor {.inline.} explicitly default to being inline, but this may change in future versions of the implementation.

The iterator type is always of the calling convention closure implicitly; the following example shows how to use iterators to implement a collaborative tasking system:

# simple tasking:
  Task = iterator (ticker: int)

iterator a1(ticker: int) {.closure.} =
  echo "a1: A"
  echo "a1: B"
  echo "a1: C"
  echo "a1: D"

iterator a2(ticker: int) {.closure.} =
  echo "a2: A"
  echo "a2: B"
  echo "a2: C"

proc runTasks(t: varargs[Task]) =
  var ticker = 0
  while true:
    let x = t[ticker mod t.len]
    if finished(x): break
    inc ticker

runTasks(a1, a2)

The builtin system.finished can be used to determine if an iterator has finished its operation; no exception is raised on an attempt to invoke an iterator that has already finished its work.

Note that system.finished is error prone to use because it only returns true one iteration after the iterator has finished:

iterator mycount(a, b: int): int {.closure.} =
  var x = a
  while x <= b:
    yield x
    inc x

var c = mycount # instantiate the iterator
while not finished(c):
  echo c(1, 3)

# Produces

Instead this code has to be used:

var c = mycount # instantiate the iterator
while true:
  let value = c(1, 3)
  if finished(c): break # and discard 'value'!
  echo value

It helps to think that the iterator actually returns a pair (value, done) and finished is used to access the hidden done field.

Closure iterators are resumable functions and so one has to provide the arguments to every call. To get around this limitation one can capture parameters of an outer factory proc:

proc mycount(a, b: int): iterator (): int =
  result = iterator (): int =
    var x = a
    while x <= b:
      yield x
      inc x

let foo = mycount(1, 4)

for f in foo():
  echo f


A converter is like an ordinary proc except that it enhances the "implicitly convertible" type relation (see Convertible relation):

# bad style ahead: Nim is not C.
converter toBool(x: int): bool = x != 0

if 4:
  echo "compiles"

A converter can also be explicitly invoked for improved readability. Note that implicit converter chaining is not supported: If there is a converter from type A to type B and from type B to type C the implicit conversion from A to C is not provided.

Type sections


type # example demonstrating mutually recursive types
  Node = ref object  # an object managed by the garbage collector (ref)
    le, ri: Node     # left and right subtrees
    sym: ref Sym     # leaves contain a reference to a Sym
  Sym = object       # a symbol
    name: string     # the symbol's name
    line: int        # the line the symbol was declared in
    code: Node       # the symbol's abstract syntax tree

A type section begins with the type keyword. It contains multiple type definitions. A type definition binds a type to a name. Type definitions can be recursive or even mutually recursive. Mutually recursive types are only possible within a single type section. Nominal types like objects or enums can only be defined in a type section.

Exception handling

Try statement


# read the first two lines of a text file that should contain numbers
# and tries to add them
  f: File
if open(f, "numbers.txt"):
    var a = readLine(f)
    var b = readLine(f)
    echo "sum: " & $(parseInt(a) + parseInt(b))
  except OverflowError:
    echo "overflow!"
  except ValueError:
    echo "could not convert string to integer"
  except IOError:
    echo "IO error!"
    echo "Unknown exception!"

The statements after the try are executed in sequential order unless an exception e is raised. If the exception type of e matches any listed in an except clause the corresponding statements are executed. The statements following the except clauses are called exception handlers.

The empty except clause is executed if there is an exception that is not listed otherwise. It is similar to an else clause in if statements.

If there is a finally clause, it is always executed after the exception handlers.

The exception is consumed in an exception handler. However, an exception handler may raise another exception. If the exception is not handled, it is propagated through the call stack. This means that often the rest of the procedure - that is not within a finally clause - is not executed (if an exception occurs).

Try expression

Try can also be used as an expression; the type of the try branch then needs to fit the types of except branches, but the type of the finally branch always has to be void:

let x = try: parseInt("133a")
        except: -1
        finally: echo "hi"

To prevent confusing code there is a parsing limitation; if the try follows a ( it has to be written as a one liner:

let x = (try: parseInt("133a") except: -1)

Except clauses

Within an except clause, it is possible to use getCurrentException to retrieve the exception that has been raised:

  # ...
except IOError:
  let e = getCurrentException()
  # Now use "e"

Note that getCurrentException always returns a ref Exception type. If a variable of the proper type is needed (in the example above, IOError), one must convert it explicitly:

  # ...
except IOError:
  let e = (ref IOError)(getCurrentException())
  # "e" is now of the proper type

However, this is seldom needed. The most common case is to extract an error message from e, and for such situations it is enough to use getCurrentExceptionMsg:

  # ...
except IOError:
  echo "I/O error: " & getCurrentExceptionMsg()

Defer statement

Instead of a try finally statement a defer statement can be used.

Any statements following the defer in the current block will be considered to be in an implicit try block:

proc main =
  var f = open("numbers.txt")
  defer: close(f)
  f.write "abc"
  f.write "def"

Is rewritten to:

proc main =
  var f = open("numbers.txt")
    f.write "abc"
    f.write "def"

Top level defer statements are not supported since it's unclear what such a statement should refer to.

Raise statement


raise newEOS("operating system failed")

Apart from built-in operations like array indexing, memory allocation, etc. the raise statement is the only way to raise an exception.

If no exception name is given, the current exception is re-raised. The ReraiseError exception is raised if there is no exception to re-raise. It follows that the raise statement always raises an exception.

Exception hierarchy

The exception tree is defined in the system module. Every exception inherits from system.Exception. Exceptions that indicate programming bugs inherit from system.Defect (which is a subtype of Exception) and are stricly speaking not catchable as they can also be mapped to an operation that terminates the whole process. Exceptions that indicate any other runtime error that can be caught inherit from system.CatchableError (which is a subtype of Exception).

Imported exceptions

It is possible to raise/catch imported C++ exceptions. Types imported using importcpp can be raised or caught. Exceptions are raised by value and caught by reference. Example:

  std_exception {.importcpp: "std::exception", header: "<exception>".} = object

proc what(s: std_exception): cstring {.importcpp: "((char *)#.what())".}

  raise std_exception()
except std_exception as ex:
  echo ex.what()

Effect system

Exception tracking

Nim supports exception tracking. The raises pragma can be used to explicitly define which exceptions a proc/iterator/method/converter is allowed to raise. The compiler verifies this:

proc p(what: bool) {.raises: [IOError, OSError].} =
  if what: raise newException(IOError, "IO")
  else: raise newException(OSError, "OS")

An empty raises list (raises: []) means that no exception may be raised:

proc p(): bool {.raises: [].} =
    result = true
    result = false

A raises list can also be attached to a proc type. This affects type compatibility:

  Callback = proc (s: string) {.raises: [IOError].}
  c: Callback

proc p(x: string) =
  raise newException(OSError, "OS")

c = p # type error

For a routine p the compiler uses inference rules to determine the set of possibly raised exceptions; the algorithm operates on p's call graph:

  1. Every indirect call via some proc type T is assumed to raise system.Exception (the base type of the exception hierarchy) and thus any exception unless T has an explicit raises list. However if the call is of the form f(...) where f is a parameter of the currently analysed routine it is ignored. The call is optimistically assumed to have no effect. Rule 2 compensates for this case.
  2. Every expression of some proc type within a call that is not a call itself (and not nil) is assumed to be called indirectly somehow and thus its raises list is added to p's raises list.
  3. Every call to a proc q which has an unknown body (due to a forward declaration or an importc pragma) is assumed to raise system.Exception unless q has an explicit raises list.
  4. Every call to a method m is assumed to raise system.Exception unless m has an explicit raises list.
  5. For every other call the analysis can determine an exact raises list.
  6. For determining a raises list, the raise and try statements of p are taken into consideration.

Rules 1-2 ensure the following works:

proc noRaise(x: proc()) {.raises: [].} =
  # unknown call that might raise anything, but valid:

proc doRaise() {.raises: [IOError].} =
  raise newException(IOError, "IO")

proc use() {.raises: [].} =
  # doesn't compile! Can raise IOError!

So in many cases a callback does not cause the compiler to be overly conservative in its effect analysis.

Tag tracking

The exception tracking is part of Nim's effect system. Raising an exception is an effect. Other effects can also be defined. A user defined effect is a means to tag a routine and to perform checks against this tag:

type IO = object ## input/output effect
proc readLine(): string {.tags: [IO].} = discard

proc no_IO_please() {.tags: [].} =
  # the compiler prevents this:
  let x = readLine()

A tag has to be a type name. A tags list - like a raises list - can also be attached to a proc type. This affects type compatibility.

The inference for tag tracking is analogous to the inference for exception tracking.

Read/Write tracking

Note: Read/write tracking is not yet implemented!

The inference for read/write tracking is analogous to the inference for exception tracking.

Effects pragma

The effects pragma has been designed to assist the programmer with the effects analysis. It is a statement that makes the compiler output all inferred effects up to the effects's position:

proc p(what: bool) =
  if what:
    raise newException(IOError, "IO")
    raise newException(OSError, "OS")

The compiler produces a hint message that IOError can be raised. OSError is not listed as it cannot be raised in the branch the effects pragma appears in.


Generics are Nim's means to parametrize procs, iterators or types with type parameters. Depending on context, the brackets are used either to introduce type parameters or to instantiate a generic proc, iterator or type.

The following example shows a generic binary tree can be modelled:

  BinaryTree*[T] = ref object # BinaryTree is a generic type with
                              # generic param ``T``
    le, ri: BinaryTree[T]     # left and right subtrees; may be nil
    data: T                   # the data stored in a node

proc newNode*[T](data: T): BinaryTree[T] =
  # constructor for a node
  result = BinaryTree[T](le: nil, ri: nil, data: data)

proc add*[T](root: var BinaryTree[T], n: BinaryTree[T]) =
  # insert a node into the tree
  if root == nil:
    root = n
    var it = root
    while it != nil:
      # compare the data items; uses the generic ``cmp`` proc
      # that works for any type that has a ``==`` and ``<`` operator
      var c = cmp(it.data, n.data)
      if c < 0:
        if it.le == nil:
          it.le = n
        it = it.le
        if it.ri == nil:
          it.ri = n
        it = it.ri

proc add*[T](root: var BinaryTree[T], data: T) =
  # convenience proc:
  add(root, newNode(data))

iterator preorder*[T](root: BinaryTree[T]): T =
  # Preorder traversal of a binary tree.
  # Since recursive iterators are not yet implemented,
  # this uses an explicit stack (which is more efficient anyway):
  var stack: seq[BinaryTree[T]] = @[root]
  while stack.len > 0:
    var n = stack.pop()
    while n != nil:
      yield n.data
      add(stack, n.ri)  # push right subtree onto the stack
      n = n.le          # and follow the left pointer

  root: BinaryTree[string] # instantiate a BinaryTree with ``string``
add(root, newNode("hello")) # instantiates ``newNode`` and ``add``
add(root, "world")          # instantiates the second ``add`` proc
for str in preorder(root):

The T is called a generic type parameter or a type variable.

Is operator

The is operator checks for type equivalence at compile time. It is therefore very useful for type specialization within generic code:

  Table[Key, Value] = object
    keys: seq[Key]
    values: seq[Value]
    when not (Key is string): # empty value for strings used for optimization
      deletedKeys: seq[bool]

Type Classes

A type class is a special pseudo-type that can be used to match against types in the context of overload resolution or the is operator. Nim supports the following built-in type classes:

type class matches
object any object type
tuple any tuple type
enum any enumeration
proc any proc type
ref any ref type
ptr any ptr type
var any var type
distinct any distinct type
array any array type
set any set type
seq any seq type
auto any type
any distinct auto (see below)

Furthermore, every generic type automatically creates a type class of the same name that will match any instantiation of the generic type.

Type classes can be combined using the standard boolean operators to form more complex type classes:

# create a type class that will match all tuple and object types
type RecordType = tuple or object

proc printFields(rec: RecordType) =
  for key, value in fieldPairs(rec):
    echo key, " = ", value

Procedures utilizing type classes in such manner are considered to be implicitly generic. They will be instantiated once for each unique combination of param types used within the program.

Nim also allows for type classes and regular types to be specified as type constraints of the generic type parameter:

proc onlyIntOrString[T: int|string](x, y: T) = discard

onlyIntOrString(450, 616) # valid
onlyIntOrString(5.0, 0.0) # type mismatch
onlyIntOrString("xy", 50) # invalid as 'T' cannot be both at the same time

By default, during overload resolution each named type class will bind to exactly one concrete type. We call such type classes bind once types. Here is an example taken directly from the system module to illustrate this:

proc `==`*(x, y: tuple): bool =
  ## requires `x` and `y` to be of the same tuple type
  ## generic ``==`` operator for tuples that is lifted from the components
  ## of `x` and `y`.
  result = true
  for a, b in fields(x, y):
    if a != b: result = false

Alternatively, the distinct type modifier can be applied to the type class to allow each param matching the type class to bind to a different type. Such type classes are called bind many types.

Procs written with the implicitly generic style will often need to refer to the type parameters of the matched generic type. They can be easily accessed using the dot syntax:

type Matrix[T, Rows, Columns] = object

proc `[]`(m: Matrix, row, col: int): Matrix.T =
  m.data[col * high(Matrix.Columns) + row]

Alternatively, the type operator can be used over the proc params for similar effect when anonymous or distinct type classes are used.

When a generic type is instantiated with a type class instead of a concrete type, this results in another more specific type class:

seq[ref object]  # Any sequence storing references to any object type

type T1 = auto
proc foo(s: seq[T1], e: T1)
  # seq[T1] is the same as just `seq`, but T1 will be allowed to bind
  # to a single type, while the signature is being matched

Matrix[Ordinal] # Any Matrix instantiation using integer values

As seen in the previous example, in such instantiations, it's not necessary to supply all type parameters of the generic type, because any missing ones will be inferred to have the equivalent of the any type class and thus they will match anything without discrimination.

Generic inference restrictions

The types var T and typedesc[T] cannot be inferred in a generic instantiation. The following is not allowed:

proc g[T](f: proc(x: T); x: T) =

proc c(y: int) = echo y
proc v(y: var int) =
  y += 100
var i: int

# allowed: infers 'T' to be of type 'int'
g(c, 42)

# not valid: 'T' is not inferred to be of type 'var int'
g(v, i)

# also not allowed: explict instantiation via 'var int'
g[var int](v, i)


Note: Concepts are still in development.

Concepts, also known as "user-defined type classes", are used to specify an arbitrary set of requirements that the matched type must satisfy.

Concepts are written in the following form:

  Comparable = concept x, y
    (x < y) is bool
  Stack[T] = concept s, var v
    s.pop() is T
    s.len is Ordinal
    for value in s:
      value is T

The concept is a match if:

  1. all of the expressions within the body can be compiled for the tested type
  2. all statically evaluable boolean expressions in the body must be true

The identifiers following the concept keyword represent instances of the currently matched type. You can apply any of the standard type modifiers such as var, ref, ptr and static to denote a more specific type of instance. You can also apply the type modifier to create a named instance of the type itself:

  MyConcept = concept x, var v, ref r, ptr p, static s, type T

Within the concept body, types can appear in positions where ordinary values and parameters are expected. This provides a more convenient way to check for the presence of callable symbols with specific signatures:

  OutputStream = concept var s

In order to check for symbols accepting type params, you must prefix the type with the explicit type modifier. The named instance of the type, following the concept keyword is also considered to have the explicit modifier and will be matched only as a type.

  # Let's imagine a user-defined casting framework with operators
  # such as `val.to(string)` and `val.to(JSonValue)`. We can test
  # for these with the following concept:
  MyCastables = concept x
    x.to(type string)
    x.to(type JSonValue)
  # Let's define a couple of concepts, known from Algebra:
  AdditiveMonoid* = concept x, y, type T
    x + y is T
    T.zero is T # require a proc such as `int.zero` or 'Position.zero'
  AdditiveGroup* = concept x, y, type T
    x is AdditiveMonoid
    -x is T
    x - y is T

Please note that the is operator allows one to easily verify the precise type signatures of the required operations, but since type inference and default parameters are still applied in the concept body, it's also possible to describe usage protocols that do not reveal implementation details.

Much like generics, concepts are instantiated exactly once for each tested type and any static code included within the body is executed only once.

Concept diagnostics

By default, the compiler will report the matching errors in concepts only when no other overload can be selected and a normal compilation error is produced. When you need to understand why the compiler is not matching a particular concept and, as a result, a wrong overload is selected, you can apply the explain pragma to either the concept body or a particular call-site.

  MyConcept {.explain.} = concept ...

overloadedProc(x, y, z) {.explain.}

This will provide Hints in the compiler output either every time the concept is not matched or only on the particular call-site.

Generic concepts and type binding rules

The concept types can be parametric just like the regular generic types:

### matrixalgo.nim

import typetraits

  AnyMatrix*[R, C: static int; T] = concept m, var mvar, type M
    M.ValueType is T
    M.Rows == R
    M.Cols == C
    m[int, int] is T
    mvar[int, int] = T
    type TransposedType = stripGenericParams(M)[C, R, T]
  AnySquareMatrix*[N: static int, T] = AnyMatrix[N, N, T]
  AnyTransform3D* = AnyMatrix[4, 4, float]

proc transposed*(m: AnyMatrix): m.TransposedType =
  for r in 0 ..< m.R:
    for c in 0 ..< m.C:
      result[r, c] = m[c, r]

proc determinant*(m: AnySquareMatrix): int =

proc setPerspectiveProjection*(m: AnyTransform3D) =

### matrix.nim

  Matrix*[M, N: static int; T] = object
    data: array[M*N, T]

proc `[]`*(M: Matrix; m, n: int): M.T =
  M.data[m * M.N + n]

proc `[]=`*(M: var Matrix; m, n: int; v: M.T) =
  M.data[m * M.N + n] = v

# Adapt the Matrix type to the concept's requirements
template Rows*(M: type Matrix): expr = M.M
template Cols*(M: type Matrix): expr = M.N
template ValueType*(M: type Matrix): type = M.T

### usage.nim

import matrix, matrixalgo

  m: Matrix[3, 3, int]
  projectionMatrix: Matrix[4, 4, float]

echo m.transposed.determinant
setPerspectiveProjection projectionMatrix

When the concept type is matched against a concrete type, the unbound type parameters are inferred from the body of the concept in a way that closely resembles the way generic parameters of callable symbols are inferred on call sites.

Unbound types can appear both as params to calls such as s.push(T) and on the right-hand side of the is operator in cases such as x.pop is T and x.data is seq[T].

Unbound static params will be inferred from expressions involving the == operator and also when types dependent on them are being matched:

  MatrixReducer[M, N: static int; T] = concept x
    x.reduce(SquareMatrix[N, T]) is array[M, int]

The Nim compiler includes a simple linear equation solver, allowing it to infer static params in some situations where integer arithmetic is involved.

Just like in regular type classes, Nim discriminates between bind once and bind many types when matching the concept. You can add the distinct modifier to any of the otherwise inferable types to get a type that will be matched without permanently inferring it. This may be useful when you need to match several procs accepting the same wide class of types:

  Enumerable[T] = concept e
    for v in e:
      v is T

  MyConcept = concept o
    # this could be inferred to a type such as Enumerable[int]
    o.foo is distinct Enumerable
    # this could be inferred to a different type such as Enumerable[float]
    o.bar is distinct Enumerable
    # it's also possible to give an alias name to a `bind many` type class
    type Enum = distinct Enumerable
    o.baz is Enum

On the other hand, using bind once types allows you to test for equivalent types used in multiple signatures, without actually requiring any concrete types, thus allowing you to encode implementation-defined types:

  MyConcept = concept x
    type T1 = auto
    x.bar(T1) # both procs must accept the same type
    type T2 = seq[SomeNumber]
    x.omega(T2) # both procs must accept the same type
                # and it must be a numeric sequence

As seen in the previous examples, you can refer to generic concepts such as Enumerable[T] just by their short name. Much like the regular generic types, the concept will be automatically instantiated with the bind once auto type in the place of each missing generic param.

Please note that generic concepts such as Enumerable[T] can be matched against concrete types such as string. Nim doesn't require the concept type to have the same number of parameters as the type being matched. If you wish to express a requirement towards the generic parameters of the matched type, you can use a type mapping operator such as genericHead or stripGenericParams within the body of the concept to obtain the uninstantiated version of the type, which you can then try to instantiate in any required way. For example, here is how one might define the classic Functor concept from Haskell and then demonstrate that Nim's Option[T] type is an instance of it:

import sugar, typetraits

  Functor[A] = concept f
    type MatchedGenericType = genericHead(f.type)
      # `f` will be a value of a type such as `Option[T]`
      # `MatchedGenericType` will become the `Option` type
    f.val is A
      # The Functor should provide a way to obtain
      # a value stored inside it
    type T = auto
    map(f, A -> T) is MatchedGenericType[T]
      # And it should provide a way to map one instance of
      # the Functor to a instance of a different type, given
      # a suitable `map` operation for the enclosed values

import options
echo Option[int] is Functor # prints true

Concept derived values

All top level constants or types appearing within the concept body are accessible through the dot operator in procs where the concept was successfully matched to a concrete type:

  DateTime = concept t1, t2, type T
    const Min = T.MinDate
    T.Now is T
    t1 < t2 is bool
    type TimeSpan = type(t1 - t2)
    TimeSpan * int is TimeSpan
    TimeSpan + TimeSpan is TimeSpan
    t1 + TimeSpan is T

proc eventsJitter(events: Enumerable[DateTime]): float =
    # this variable will have the inferred TimeSpan type for
    # the concrete Date-like value the proc was called with:
    averageInterval: DateTime.TimeSpan
    deviation: float

Concept refinement

When the matched type within a concept is directly tested against a different concept, we say that the outer concept is a refinement of the inner concept and thus it is more-specific. When both concepts are matched in a call during overload resolution, Nim will assign a higher precedence to the most specific one. As an alternative way of defining concept refinements, you can use the object inheritance syntax involving the of keyword:

  Graph = concept g, type G of EqualyComparable, Copyable
      VertexType = G.VertexType
      EdgeType = G.EdgeType
    VertexType is Copyable
    EdgeType is Copyable
      v: VertexType
      e: EdgeType
  IncidendeGraph = concept of Graph
    # symbols such as variables and types from the refined
    # concept are automatically in scope:
    g.source(e) is VertexType
    g.target(e) is VertexType
    g.outgoingEdges(v) is Enumerable[EdgeType]
  BidirectionalGraph = concept g, type G
    # The following will also turn the concept into a refinement when it
    # comes to overload resolution, but it doesn't provide the convenient
    # symbol inheritance
    g is IncidendeGraph
    g.incomingEdges(G.VertexType) is Enumerable[G.EdgeType]

proc f(g: IncidendeGraph)
proc f(g: BidirectionalGraph) # this one will be preferred if we pass a type
                              # matching the BidirectionalGraph concept

Symbol lookup in generics

Open and Closed symbols

The symbol binding rules in generics are slightly subtle: There are "open" and "closed" symbols. A "closed" symbol cannot be re-bound in the instantiation context, an "open" symbol can. Per default overloaded symbols are open and every other symbol is closed.

Open symbols are looked up in two different contexts: Both the context at definition and the context at instantiation are considered:

  Index = distinct int

proc `==` (a, b: Index): bool {.borrow.}

var a = (0, 0.Index)
var b = (0, 0.Index)

echo a == b # works!

In the example the generic == for tuples (as defined in the system module) uses the == operators of the tuple's components. However, the == for the Index type is defined after the == for tuples; yet the example compiles as the instantiation takes the currently defined symbols into account too.

Mixin statement

A symbol can be forced to be open by a mixin declaration:

proc create*[T](): ref T =
  # there is no overloaded 'init' here, so we need to state that it's an
  # open symbol explicitly:
  mixin init
  new result
  init result

Bind statement

The bind statement is the counterpart to the mixin statement. It can be used to explicitly declare identifiers that should be bound early (i.e. the identifiers should be looked up in the scope of the template/generic definition):

# Module A
  lastId = 0

template genId*: untyped =
  bind lastId
# Module B
import A

echo genId()

But a bind is rarely useful because symbol binding from the definition scope is the default.


A template is a simple form of a macro: It is a simple substitution mechanism that operates on Nim's abstract syntax trees. It is processed in the semantic pass of the compiler.

The syntax to invoke a template is the same as calling a procedure.


template `!=` (a, b: untyped): untyped =
  # this definition exists in the System module
  not (a == b)

assert(5 != 6) # the compiler rewrites that to: assert(not (5 == 6))

The !=, >, >=, in, notin, isnot operators are in fact templates:

a > b is transformed into b < a.
a in b is transformed into contains(b, a).
notin and isnot have the obvious meanings.

The "types" of templates can be the symbols untyped, typed or type. These are "meta types", they can only be used in certain contexts. Regular types can be used too; this implies that typed expressions are expected.

Typed vs untyped parameters

An untyped parameter means that symbol lookups and type resolution is not performed before the expression is passed to the template. This means that for example undeclared identifiers can be passed to the template:

template declareInt(x: untyped) =
  var x: int

declareInt(x) # valid
x = 3
template declareInt(x: typed) =
  var x: int

declareInt(x) # invalid, because x has not been declared and so has no type

A template where every parameter is untyped is called an immediate template. For historical reasons templates can be explicitly annotated with an immediate pragma and then these templates do not take part in overloading resolution and the parameters' types are ignored by the compiler. Explicit immediate templates are now deprecated.

Note: For historical reasons stmt is an alias for typed and expr an alias for untyped, but new code should use the newer, clearer names.

Passing a code block to a template

You can pass a block of statements as the last argument to a template following the special : syntax:

template withFile(f, fn, mode, actions: untyped): untyped =
  var f: File
  if open(f, fn, mode):
    quit("cannot open: " & fn)

withFile(txt, "ttempl3.txt", fmWrite):  # special colon
  txt.writeLine("line 1")
  txt.writeLine("line 2")

In the example, the two writeLine statements are bound to the actions parameter.

Usually to pass a block of code to a template the parameter that accepts the block needs to be of type untyped. Because symbol lookups are then delayed until template instantiation time:

template t(body: typed) =

  var i = 1
  echo i

  var i = 2  # fails with 'attempt to redeclare i'
  echo i

The above code fails with the mysterious error message that i has already been declared. The reason for this is that the var i = ... bodies need to be type-checked before they are passed to the body parameter and type checking in Nim implies symbol lookups. For the symbol lookups to succeed i needs to be added to the current (i.e. outer) scope. After type checking these additions to the symbol table are not rolled back (for better or worse). The same code works with untyped as the passed body is not required to be type-checked:

template t(body: untyped) =

  var i = 1
  echo i

  var i = 2  # compiles
  echo i

Varargs of untyped

In addition to the untyped meta-type that prevents type checking there is also varargs[untyped] so that not even the number of parameters is fixed:

template hideIdentifiers(x: varargs[untyped]) = discard

hideIdentifiers(undeclared1, undeclared2)

However, since a template cannot iterate over varargs, this feature is generally much more useful for macros.

Note: For historical reasons varargs[expr] is not equivalent to varargs[untyped].

Symbol binding in templates

A template is a hygienic macro and so opens a new scope. Most symbols are bound from the definition scope of the template:

# Module A
  lastId = 0

template genId*: untyped =
# Module B
import A

echo genId() # Works as 'lastId' has been bound in 'genId's defining scope

As in generics symbol binding can be influenced via mixin or bind statements.

Identifier construction

In templates identifiers can be constructed with the backticks notation:

template typedef(name: untyped, typ: type) =
    `T name`* {.inject.} = typ
    `P name`* {.inject.} = ref `T name`

typedef(myint, int)
var x: PMyInt

In the example name is instantiated with myint, so `T name` becomes Tmyint.

Lookup rules for template parameters

A parameter p in a template is even substituted in the expression x.p. Thus template arguments can be used as field names and a global symbol can be shadowed by the same argument name even when fully qualified:

# module 'm'

  Lev = enum
    levA, levB

var abclev = levB

template tstLev(abclev: Lev) =
  echo abclev, " ", m.abclev

# produces: 'levA levA'

But the global symbol can properly be captured by a bind statement:

# module 'm'

  Lev = enum
    levA, levB

var abclev = levB

template tstLev(abclev: Lev) =
  bind m.abclev
  echo abclev, " ", m.abclev

# produces: 'levA levB'

Hygiene in templates

Per default templates are hygienic: Local identifiers declared in a template cannot be accessed in the instantiation context:

template newException*(exceptn: type, message: string): untyped =
    e: ref exceptn  # e is implicitly gensym'ed here
  e.msg = message

# so this works:
let e = "message"
raise newException(IoError, e)

Whether a symbol that is declared in a template is exposed to the instantiation scope is controlled by the inject and gensym pragmas: gensym'ed symbols are not exposed but inject'ed are.

The default for symbols of entity type, var, let and const is gensym and for proc, iterator, converter, template, macro is inject. However, if the name of the entity is passed as a template parameter, it is an inject'ed symbol:

template withFile(f, fn, mode: untyped, actions: untyped): untyped =
    var f: File  # since 'f' is a template param, it's injected implicitly

withFile(txt, "ttempl3.txt", fmWrite):
  txt.writeLine("line 1")
  txt.writeLine("line 2")

The inject and gensym pragmas are second class annotations; they have no semantics outside of a template definition and cannot be abstracted over:

{.pragma myInject: inject.}

template t() =
  var x {.myInject.}: int # does NOT work

To get rid of hygiene in templates, one can use the dirty pragma for a template. inject and gensym have no effect in dirty templates.

Limitations of the method call syntax

The expression x in x.f needs to be semantically checked (that means symbol lookup and type checking) before it can be decided that it needs to be rewritten to f(x). Therefore the dot syntax has some limitations when it is used to invoke templates/macros:

template declareVar(name: untyped) =
  const name {.inject.} = 45

# Doesn't compile:

Another common example is this:

from sequtils import toSeq

iterator something: string =
  yield "Hello"
  yield "World"

var info = something().toSeq

The problem here is that the compiler already decided that something() as an iterator is not callable in this context before toSeq gets its chance to convert it into a sequence.


A macro is a special function that is executed at compile-time. Normally the input for a macro is an abstract syntax tree (AST) of the code that is passed to it. The macro can then do transformations on it and return the transformed AST. The transformed AST is then passed to the compiler as if the macro invocation would have been replaced by its result in the source code. This can be used to implement domain specific languages.

While macros enable advanced compile-time code transformations, they cannot change Nim's syntax. However, this is no real restriction because Nim's syntax is flexible enough anyway.

To write macros, one needs to know how the Nim concrete syntax is converted to an AST.

There are two ways to invoke a macro:

  1. invoking a macro like a procedure call (expression macros)
  2. invoking a macro with the special macrostmt syntax (statement macros)

Expression Macros

The following example implements a powerful debug command that accepts a variable number of arguments:

# to work with Nim syntax trees, we need an API that is defined in the
# ``macros`` module:
import macros

macro debug(args: varargs[untyped]): untyped =
  # `args` is a collection of `NimNode` values that each contain the
  # AST for an argument of the macro. A macro always has to
  # return a `NimNode`. A node of kind `nnkStmtList` is suitable for
  # this use case.
  result = nnkStmtList.newTree()
  # iterate over any argument that is passed to this macro:
  for n in args:
    # add a call to the statement list that writes the expression;
    # `toStrLit` converts an AST to its string representation:
    result.add newCall("write", newIdentNode("stdout"), newLit(n.repr))
    # add a call to the statement list that writes ": "
    result.add newCall("write", newIdentNode("stdout"), newLit(": "))
    # add a call to the statement list that writes the expressions value:
    result.add newCall("writeLine", newIdentNode("stdout"), n)

  a: array[0..10, int]
  x = "some string"
a[0] = 42
a[1] = 45

debug(a[0], a[1], x)

The macro call expands to:

write(stdout, "a[0]")
write(stdout, ": ")
writeLine(stdout, a[0])

write(stdout, "a[1]")
write(stdout, ": ")
writeLine(stdout, a[1])

write(stdout, "x")
write(stdout, ": ")
writeLine(stdout, x)

Arguments that are passed to a varargs parameter are wrapped in an array constructor expression. This is why debug iterates over all of n's children.


The above debug macro relies on the fact that write, writeLine and stdout are declared in the system module and thus visible in the instantiating context. There is a way to use bound identifiers (aka symbols) instead of using unbound identifiers. The bindSym builtin can be used for that:

import macros

macro debug(n: varargs[typed]): untyped =
  result = newNimNode(nnkStmtList, n)
  for x in n:
    # we can bind symbols in scope via 'bindSym':
    add(result, newCall(bindSym"write", bindSym"stdout", toStrLit(x)))
    add(result, newCall(bindSym"write", bindSym"stdout", newStrLitNode(": ")))
    add(result, newCall(bindSym"writeLine", bindSym"stdout", x))

  a: array[0..10, int]
  x = "some string"
a[0] = 42
a[1] = 45

debug(a[0], a[1], x)

The macro call expands to:

write(stdout, "a[0]")
write(stdout, ": ")
writeLine(stdout, a[0])

write(stdout, "a[1]")
write(stdout, ": ")
writeLine(stdout, a[1])

write(stdout, "x")
write(stdout, ": ")
writeLine(stdout, x)

However, the symbols write, writeLine and stdout are already bound and are not looked up again. As the example shows, bindSym does work with overloaded symbols implicitly.

Statement Macros

Statement macros are defined just as expression macros. However, they are invoked by an expression following a colon.

The following example outlines a macro that generates a lexical analyzer from regular expressions:

import macros

macro case_token(n: untyped): untyped =
  # creates a lexical analyzer from regular expressions
  # ... (implementation is an exercise for the reader :-)

case_token: # this colon tells the parser it is a macro statement
of r"[A-Za-z_]+[A-Za-z_0-9]*":
  return tkIdentifier
of r"0-9+":
  return tkInteger
of r"[\+\-\*\?]+":
  return tkOperator
  return tkUnknown

Style note: For code readability, it is the best idea to use the least powerful programming construct that still suffices. So the "check list" is:

  1. Use an ordinary proc/iterator, if possible.
  2. Else: Use a generic proc/iterator, if possible.
  3. Else: Use a template, if possible.
  4. Else: Use a macro.

Macros as pragmas

Whole routines (procs, iterators etc.) can also be passed to a template or a macro via the pragma notation:

template m(s: untyped) = discard

proc p() {.m.} = discard

This is a simple syntactic transformation into:

template m(s: untyped) = discard

  proc p() = discard

For loop macros

A macro that takes as its only input parameter an expression of the special type system.ForLoopStmt can rewrite the entirety of a for loop:

import macros
{.experimental: "forLoopMacros".}

macro enumerate(x: ForLoopStmt): untyped =
  expectKind x, nnkForStmt
  # we strip off the first for loop variable and use
  # it as an integer counter:
  result = newStmtList()
  result.add newVarStmt(x[0], newLit(0))
  var body = x[^1]
  if body.kind != nnkStmtList:
    body = newTree(nnkStmtList, body)
  body.add newCall(bindSym"inc", x[0])
  var newFor = newTree(nnkForStmt)
  for i in 1..x.len-3:
    newFor.add x[i]
  # transform enumerate(X) to 'X'
  newFor.add x[^2][1]
  newFor.add body
  result.add newFor
  # now wrap the whole macro in a block to create a new scope
  result = quote do:
    block: `result`

for a, b in enumerate(items([1, 2, 3])):
  echo a, " ", b

# without wrapping the macro in a block, we'd need to choose different
# names for `a` and `b` here to avoid redefinition errors
for a, b in enumerate([1, 2, 3, 5]):
  echo a, " ", b

Currently for loop macros must be enabled explicitly via {.experimental: "forLoopMacros".}.

Case statement macros

A macro that needs to be called match can be used to rewrite case statements in order to implement pattern matching for certain types. The following example implements a simplistic form of pattern matching for tuples, leveraging the existing equality operator for tuples (as provided in system.==):

{.experimental: "caseStmtMacros".}

import macros

macro match(n: tuple): untyped =
  result = newTree(nnkIfStmt)
  let selector = n[0]
  for i in 1 ..< n.len:
    let it = n[i]
    case it.kind
    of nnkElse, nnkElifBranch, nnkElifExpr, nnkElseExpr:
      result.add it
    of nnkOfBranch:
      for j in 0..it.len-2:
        let cond = newCall("==", selector, it[j])
        result.add newTree(nnkElifBranch, cond, it[^1])
      error "'match' cannot handle this node", it
  echo repr result

case ("foo", 78)
of ("foo", 78): echo "yes"
of ("bar", 88): echo "no"
else: discard

Currently case statement macros must be enabled explicitly via {.experimental: "caseStmtMacros".}.

match macros are subject to overload resolution. First the case's selector expression is used to determine which match macro to call. To this macro is then passed the complete case statement body and the macro is evaluated.

In other words, the macro needs to transform the full case statement but only the statement's selector expression is used to determine which macro to call.

Special Types


Note: static[T] is still in development.

As their name suggests, static parameters must be known at compile-time:

proc precompiledRegex(pattern: static string): RegEx =
  var res {.global.} = re(pattern)
  return res

precompiledRegex("/d+") # Replaces the call with a precompiled
                        # regex, stored in a global variable

precompiledRegex(paramStr(1)) # Error, command-line options
                              # are not known at compile-time

For the purposes of code generation, all static params are treated as generic params - the proc will be compiled separately for each unique supplied value (or combination of values).

Static params can also appear in the signatures of generic types:

  Matrix[M,N: static int; T: Number] = array[0..(M*N - 1), T]
    # Note how `Number` is just a type constraint here, while
    # `static int` requires us to supply a compile-time int value
  AffineTransform2D[T] = Matrix[3, 3, T]
  AffineTransform3D[T] = Matrix[4, 4, T]

var m1: AffineTransform3D[float]  # OK
var m2: AffineTransform2D[string] # Error, `string` is not a `Number`

Please note that static T is just a syntactic convenience for the underlying generic type static[T]. The type param can be omitted to obtain the type class of all values known at compile-time. A more specific type class can be created by instantiating static with another type class.

You can force the evaluation of a certain expression at compile-time by coercing it to a corresponding static type:

import math

echo static(fac(5)), " ", static[bool](16.isPowerOfTwo)

The complier will report any failure to evaluate the expression or a possible type mismatch error.


In many contexts, Nim allows you to treat the names of types as regular values. These values exists only during the compilation phase, but since all values must have a type, type is considered their special type.

type acts like a generic type. For instance, the type of the symbol int is type[int]. Just like with regular generic types, when the generic param is ommited, type denotes the type class of all types. As a syntactic convenience, you can also use type as a modifier. type int is considered the same as type[int].

Procs featuring type params are considered implicitly generic. They will be instantiated for each unique combination of supplied types and within the body of the proc, the name of each param will refer to the bound concrete type:

proc new(T: type): ref T =
  echo "allocating ", T.name

var n = Node.new
var tree = new(BinaryTree[int])

When multiple type params are present, they will bind freely to different types. To force a bind-once behavior one can use an explicit generic param:

proc acceptOnlyTypePairs[T, U](A, B: type[T]; C, D: type[U])

Once bound, type params can appear in the rest of the proc signature:

template declareVariableWithType(T: type, value: T) =
  var x: T = value

declareVariableWithType int, 42

Overload resolution can be further influenced by constraining the set of types that will match the type param:

template maxval(T: type int): int = high(int)
template maxval(T: type float): float = Inf

var i = int.maxval
var f = float.maxval
when false:
  var s = string.maxval # error, maxval is not implemented for string

The constraint can be a concrete type or a type class.

type operator

You can obtain the type of a given expression by constructing a type value from it (in many other languages this is known as the typeof operator):

var x = 0
var y: type(x) # y has type int

You may add a constraint to the resulting type to trigger a compile-time error if the expression doesn't have the expected type:

var x = 0
var y: type[object](x) # Error: type mismatch: got <int> but expected 'object'

If type is used to determine the result type of a proc/iterator/converter call c(X) (where X stands for a possibly empty list of arguments), the interpretation where c is an iterator is preferred over the other interpretations:

import strutils

# strutils contains both a ``split`` proc and iterator, but since an
# an iterator is the preferred interpretation, `y` has the type ``string``:
var y: type("a b c".split)

Special Operators

dot operators

Note: Dot operators are still experimental and so need to be enabled via {.experimental: "dotOperators".}.

Nim offers a special family of dot operators that can be used to intercept and rewrite proc call and field access attempts, referring to previously undeclared symbol names. They can be used to provide a fluent interface to objects lying outside the static confines of the type system such as values from dynamic scripting languages or dynamic file formats such as JSON or XML.

When Nim encounters an expression that cannot be resolved by the standard overload resolution rules, the current scope will be searched for a dot operator that can be matched against a re-written form of the expression, where the unknown field or proc name is passed to an untyped parameter:

a.b # becomes `.`(a, b)
a.b(c, d) # becomes `.`(a, b, c, d)

The matched dot operators can be symbols of any callable kind (procs, templates and macros), depending on the desired effect:

template `.` (js: PJsonNode, field: untyped): JSON = js[astToStr(field)]

var js = parseJson("{ x: 1, y: 2}")
echo js.x # outputs 1
echo js.y # outputs 2

The following dot operators are available:

operator .

This operator will be matched against both field accesses and method calls.

operator .()

This operator will be matched exclusively against method calls. It has higher precedence than the . operator and this allows one to handle expressions like x.y and x.y() differently if one is interfacing with a scripting language for example.

operator .=

This operator will be matched against assignments to missing fields.

a.b = c # becomes `.=`(a, b, c)

Type bound operations

There are 3 operations that are bound to a type:

  1. Assignment
  2. Destruction
  3. Deep copying for communication between threads

These operations can be overridden instead of overloaded. This means the implementation is automatically lifted to structured types. For instance if type T has an overridden assignment operator = this operator is also used for assignments of the type seq[T]. Since these operations are bound to a type they have to be bound to a nominal type for reasons of simplicity of implementation: This means an overridden deepCopy for ref T is really bound to T and not to ref T. This also means that one cannot override deepCopy for both ptr T and ref T at the same time; instead a helper distinct or object type has to be used for one pointer type.

operator =

This operator is the assignment operator. Note that in the contexts result = expr, parameter = defaultValue or for parameter passing no assignment is performed. For a type T that has an overloaded assignment operator var v = T() is rewritten to var v: T; v = T(); in other words var and let contexts do count as assignments.

The assignment operator needs to be attached to an object or distinct type T. Its signature has to be (var T, T). Example:

  Concrete = object
    a, b: string

proc `=`(d: var Concrete; src: Concrete) =
  shallowCopy(d.a, src.a)
  shallowCopy(d.b, src.b)
  echo "Concrete '=' called"

var x, y: array[0..2, Concrete]
var cA, cB: Concrete

var cATup, cBTup: tuple[x: int, ha: Concrete]

x = y
cA = cB
cATup = cBTup


A destructor must have a single parameter with a concrete type (the name of a generic type is allowed too). The name of the destructor has to be =destroy.

=destroy(v) will be automatically invoked for every local stack variable v that goes out of scope.

If a structured type features a field with destructable type and the user has not provided an explicit implementation, a destructor for the structured type will be automatically generated. Calls to any base class destructors in both user-defined and generated destructors will be inserted.

A destructor is attached to the type it destructs; expressions of this type can then only be used in destructible contexts and as parameters:

  MyObj = object
    x, y: int
    p: pointer

proc `=destroy`(o: var MyObj) =
  if o.p != nil: dealloc o.p

proc open: MyObj =
  result = MyObj(x: 1, y: 2, p: alloc(3))

proc work(o: MyObj) =
  echo o.x
  # No destructor invoked here for 'o' as 'o' is a parameter.

proc main() =
  # destructor automatically invoked at the end of the scope:
  var x = open()
  # valid: pass 'x' to some other proc:
  # Error: usage of a type with a destructor in a non destructible context
  echo open()

A destructible context is currently only the following:

  1. The expr in var x = expr.
  2. The expr in let x = expr.
  3. The expr in return expr.
  4. The expr in result = expr where result is the special symbol introduced by the compiler.

These rules ensure that the construction is tied to a variable and can easily be destructed at its scope exit. Later versions of the language will improve the support of destructors.

Be aware that destructors are not called for objects allocated with new. This may change in future versions of language, but for now the finalizer parameter to new has to be used.

Note: Destructors are still experimental and the spec might change significantly in order to incorporate an escape analysis.


=deepCopy is a builtin that is invoked whenever data is passed to a spawn'ed proc to ensure memory safety. The programmer can override its behaviour for a specific ref or ptr type T. (Later versions of the language may weaken this restriction.)

The signature has to be:

proc `=deepCopy`(x: T): T

This mechanism will be used by most data structures that support shared memory like channels to implement thread safe automatic memory management.

The builtin deepCopy can even clone closures and their environments. See the documentation of spawn for details.

Term rewriting macros

Term rewriting macros are macros or templates that have not only a name but also a pattern that is searched for after the semantic checking phase of the compiler: This means they provide an easy way to enhance the compilation pipeline with user defined optimizations:

template optMul{`*`(a, 2)}(a: int): int = a+a

let x = 3
echo x * 2

The compiler now rewrites x * 2 as x + x. The code inside the curlies is the pattern to match against. The operators *, **, |, ~ have a special meaning in patterns if they are written in infix notation, so to match verbatim against * the ordinary function call syntax needs to be used.

Unfortunately optimizations are hard to get right and even the tiny example is wrong:

template optMul{`*`(a, 2)}(a: int): int = a+a

proc f(): int =
  echo "side effect!"
  result = 55

echo f() * 2

We cannot duplicate 'a' if it denotes an expression that has a side effect! Fortunately Nim supports side effect analysis:

template optMul{`*`(a, 2)}(a: int{noSideEffect}): int = a+a

proc f(): int =
  echo "side effect!"
  result = 55

echo f() * 2 # not optimized ;-)

You can make one overload matching with a constraint and one without, and the one with a constraint will have precedence, and so you can handle both cases differently.

So what about 2 * a? We should tell the compiler * is commutative. We cannot really do that however as the following code only swaps arguments blindly:

template mulIsCommutative{`*`(a, b)}(a, b: int): int = b*a

What optimizers really need to do is a canonicalization:

template canonMul{`*`(a, b)}(a: int{lit}, b: int): int = b*a

The int{lit} parameter pattern matches against an expression of type int, but only if it's a literal.

Parameter constraints

The parameter constraint expression can use the operators | (or), & (and) and ~ (not) and the following predicates:

Predicate Meaning
atom The matching node has no children.
lit The matching node is a literal like "abc", 12.
sym The matching node must be a symbol (a bound identifier).
ident The matching node must be an identifier (an unbound identifier).
call The matching AST must be a call/apply expression.
lvalue The matching AST must be an lvalue.
sideeffect The matching AST must have a side effect.
nosideeffect The matching AST must have no side effect.
param A symbol which is a parameter.
genericparam A symbol which is a generic parameter.
module A symbol which is a module.
type A symbol which is a type.
var A symbol which is a variable.
let A symbol which is a let variable.
const A symbol which is a constant.
result The special result variable.
proc A symbol which is a proc.
method A symbol which is a method.
iterator A symbol which is an iterator.
converter A symbol which is a converter.
macro A symbol which is a macro.
template A symbol which is a template.
field A symbol which is a field in a tuple or an object.
enumfield A symbol which is a field in an enumeration.
forvar A for loop variable.
label A label (used in block statements).
nk* The matching AST must have the specified kind. (Example: nkIfStmt denotes an if statement.)
alias States that the marked parameter needs to alias with some other parameter.
noalias States that every other parameter must not alias with the marked parameter.

Predicates that share their name with a keyword have to be escaped with backticks: `` const . The ``alias and noalias predicates refer not only to the matching AST, but also to every other bound parameter; syntactically they need to occur after the ordinary AST predicates:

template ex{a = b + c}(a: int{noalias}, b, c: int) =
  # this transformation is only valid if 'b' and 'c' do not alias 'a':
  a = b
  inc a, c

Pattern operators

The operators *, **, |, ~ have a special meaning in patterns if they are written in infix notation.

The | operator

The | operator if used as infix operator creates an ordered choice:

template t{0|1}(): untyped = 3
let a = 1
# outputs 3:
echo a

The matching is performed after the compiler performed some optimizations like constant folding, so the following does not work:

template t{0|1}(): untyped = 3
# outputs 1:
echo 1

The reason is that the compiler already transformed the 1 into "1" for the echo statement. However, a term rewriting macro should not change the semantics anyway. In fact they can be deactivated with the --patterns:off command line option or temporarily with the patterns pragma.

The {} operator

A pattern expression can be bound to a pattern parameter via the expr{param} notation:

template t{(0|1|2){x}}(x: untyped): untyped = x+1
let a = 1
# outputs 2:
echo a

The ~ operator

The ~ operator is the not operator in patterns:

template t{x = (~x){y} and (~x){z}}(x, y, z: bool) =
  x = y
  if x: x = z

  a = false
  b = true
  c = false
a = b and c
echo a

The * operator

The * operator can flatten a nested binary expression like a & b & c to &(a, b, c):

  calls = 0

proc `&&`(s: varargs[string]): string =
  result = s[0]
  for i in 1..len(s)-1: result.add s[i]
  inc calls

template optConc{ `&&` * a }(a: string): untyped = &&a

let space = " "
echo "my" && (space & "awe" && "some " ) && "concat"

# check that it's been optimized properly:
doAssert calls == 1

The second operator of * must be a parameter; it is used to gather all the arguments. The expression "my" && (space & "awe" && "some " ) && "concat" is passed to optConc in a as a special list (of kind nkArgList) which is flattened into a call expression; thus the invocation of optConc produces:

`&&`("my", space & "awe", "some ", "concat")

The ** operator

The ** is much like the * operator, except that it gathers not only all the arguments, but also the matched operators in reverse polish notation:

import macros

  Matrix = object
    dummy: int

proc `*`(a, b: Matrix): Matrix = discard
proc `+`(a, b: Matrix): Matrix = discard
proc `-`(a, b: Matrix): Matrix = discard
proc `$`(a: Matrix): string = result = $a.dummy
proc mat21(): Matrix =
  result.dummy = 21

macro optM{ (`+`|`-`|`*`) ** a }(a: Matrix): untyped =
  echo treeRepr(a)
  result = newCall(bindSym"mat21")

var x, y, z: Matrix

echo x + y * z - x

This passes the expression x + y * z - x to the optM macro as an nnkArgList node containing:

  Sym "x"
  Sym "y"
  Sym "z"
  Sym "*"
  Sym "+"
  Sym "x"
  Sym "-"

(Which is the reverse polish notation of x + y * z - x.)


Parameters in a pattern are type checked in the matching process. If a parameter is of the type varargs it is treated specially and it can match 0 or more arguments in the AST to be matched against:

template optWrite{
  write(f, x)
  ((write|writeLine){w})(f, y)
}(x, y: varargs[untyped], f: File, w: untyped) =
  w(f, x, y)

Example: Partial evaluation

The following example shows how some simple partial evaluation can be implemented with term rewriting:

proc p(x, y: int; cond: bool): int =
  result = if cond: x + y else: x - y

template optP1{p(x, y, true)}(x, y: untyped): untyped = x + y
template optP2{p(x, y, false)}(x, y: untyped): untyped = x - y

Example: Hoisting

The following example shows how some form of hoisting can be implemented:

import pegs

template optPeg{peg(pattern)}(pattern: string{lit}): Peg =
  var gl {.global, gensym.} = peg(pattern)

for i in 0 .. 3:
  echo match("(a b c)", peg"'(' @ ')'")
  echo match("W_HI_Le", peg"\y 'while'")

The optPeg template optimizes the case of a peg constructor with a string literal, so that the pattern will only be parsed once at program startup and stored in a global gl which is then re-used. This optimization is called hoisting because it is comparable to classical loop hoisting.

AST based overloading

Parameter constraints can also be used for ordinary routine parameters; these constraints affect ordinary overloading resolution then:

proc optLit(a: string{lit|`const`}) =
  echo "string literal"
proc optLit(a: string) =
  echo "no string literal"

  constant = "abc"

  variable = "xyz"


However, the constraints alias and noalias are not available in ordinary routines.

Move optimization

The call constraint is particularly useful to implement a move optimization for types that have copying semantics:

proc `[]=`*(t: var Table, key: string, val: string) =
  ## puts a (key, value)-pair into `t`. The semantics of string require
  ## a copy here:
  let idx = findInsertionPosition(key)
  t[idx].key = key
  t[idx].val = val

proc `[]=`*(t: var Table, key: string{call}, val: string{call}) =
  ## puts a (key, value)-pair into `t`. Optimized version that knows that
  ## the strings are unique and thus don't need to be copied:
  let idx = findInsertionPosition(key)
  shallowCopy t[idx].key, key
  shallowCopy t[idx].val, val

var t: Table
# overloading resolution ensures that the optimized []= is called here:
t[f()] = g()


Nim supports splitting a program into pieces by a module concept. Each module needs to be in its own file and has its own namespace. Modules enable information hiding and separate compilation. A module may gain access to symbols of another module by the import statement. Recursive module dependencies are allowed, but slightly subtle. Only top-level symbols that are marked with an asterisk (*) are exported. A valid module name can only be a valid Nim identifier (and thus its filename is identifier.nim).

The algorithm for compiling modules is:

  • compile the whole module as usual, following import statements recursively
  • if there is a cycle only import the already parsed symbols (that are exported); if an unknown identifier occurs then abort

This is best illustrated by an example:

# Module A
  T1* = int  # Module A exports the type ``T1``
import B     # the compiler starts parsing B

proc main() =
  var i = p(3) # works because B has been parsed completely here

# Module B
import A  # A is not parsed here! Only the already known symbols
          # of A are imported.

proc p*(x: A.T1): A.T1 =
  # this works because the compiler has already
  # added T1 to A's interface symbol table
  result = x + 1

Import statement

After the import statement a list of module names can follow or a single module name followed by an except list to prevent some symbols to be imported:

import strutils except `%`, toUpperAscii

# doesn't work then:
echo "$1" % "abc".toUpperAscii

It is not checked that the except list is really exported from the module. This feature allows to compile against an older version of the module that does not export these identifiers.

Include statement

The include statement does something fundamentally different than importing a module: it merely includes the contents of a file. The include statement is useful to split up a large module into several files:

include fileA, fileB, fileC

Module names in imports

A module alias can be introduced via the as keyword:

import strutils as su, sequtils as qu

echo su.format("$1", "lalelu")

The original module name is then not accessible. The notations path/to/module or "path/to/module" can be used to refer to a module in subdirectories:

import lib/pure/os, "lib/pure/times"

Note that the module name is still strutils and not lib/pure/strutils and so one cannot do:

import lib/pure/strutils
echo lib/pure/strutils.toUpperAscii("abc")

Likewise the following does not make sense as the name is strutils already:

import lib/pure/strutils as strutils

Collective imports from a directory

The syntax import dir / [moduleA, moduleB] can be used to import multiple modules from the same directory.

Path names are syntactically either Nim identifiers or string literals. If the path name is not a valid Nim identifier it needs to be a string literal:

import "gfx/3d/somemodule" # in quotes because '3d' is not a valid Nim identifier

Pseudo import/include paths

A directory can also be a so called "pseudo directory". They can be used to avoid ambiguity when there are multiple modules with the same path.

There are two pseudo directories:

1. std: The std pseudo directory is the abstract location of Nim's standard library. For example, the syntax import std / strutils is used to unambiguously refer to the standard library's strutils module.

2. pkg: The pkg pseudo directory is used to unambiguously refer to a Nimble package. However, for technical details that lie outside of the scope of this document its semantics are: Use the search path to look for module name but ignore the standard library locations. In other words, it is the opposite of std.

From import statement

After the from statement a module name follows followed by an import to list the symbols one likes to use without explicit full qualification:

from strutils import `%`

echo "$1" % "abc"
# always possible: full qualification:
echo strutils.replace("abc", "a", "z")

It's also possible to use from module import nil if one wants to import the module but wants to enforce fully qualified access to every symbol in module.

Export statement

An export statement can be used for symbol forwarding so that client modules don't need to import a module's dependencies:

# module B
type MyObject* = object
# module A
import B
export B.MyObject

proc `$`*(x: MyObject): string = "my object"
# module C
import A

# B.MyObject has been imported implicitly here:
var x: MyObject
echo $x

When the exported symbol is another module, all of its definitions will be forwarded. You can use an except list to exclude some of the symbols.

Scope rules

Identifiers are valid from the point of their declaration until the end of the block in which the declaration occurred. The range where the identifier is known is the scope of the identifier. The exact scope of an identifier depends on the way it was declared.

Block scope

The scope of a variable declared in the declaration part of a block is valid from the point of declaration until the end of the block. If a block contains a second block, in which the identifier is redeclared, then inside this block, the second declaration will be valid. Upon leaving the inner block, the first declaration is valid again. An identifier cannot be redefined in the same block, except if valid for procedure or iterator overloading purposes.

Tuple or object scope

The field identifiers inside a tuple or object definition are valid in the following places:

  • To the end of the tuple/object definition.
  • Field designators of a variable of the given tuple/object type.
  • In all descendant types of the object type.

Module scope

All identifiers of a module are valid from the point of declaration until the end of the module. Identifiers from indirectly dependent modules are not available. The system module is automatically imported in every module.

If a module imports an identifier by two different modules, each occurrence of the identifier has to be qualified, unless it is an overloaded procedure or iterator in which case the overloading resolution takes place:

# Module A
var x*: string
# Module B
var x*: int
# Module C
import A, B
write(stdout, x) # error: x is ambiguous
write(stdout, A.x) # no error: qualifier used

var x = 4
write(stdout, x) # not ambiguous: uses the module C's x

Code reordering

Note: Code reordering is experimental and must be enabled via the {.experimental.} pragma.

The code reordering feature can implicitly rearrange procedure, template, and macro definitions along with variable declarations and initializations at the top level scope so that, to a large extent, a programmer should not have to worry about ordering definitions correctly or be forced to use forward declarations to preface definitions inside a module.


{.experimental: "codeReordering".}

proc foo(x: int) =

proc bar(x: int) =


Variables can also be reordered as well. Variables that are initialized (i.e. variables that have their declaration and assignment combined in a single statement) can have their entire initialization statement reordered. Be wary of what code is executed at the top level:

{.experimental: "codeReordering".}

proc a() =

var foo = 5

a() # outputs: "5"

It is important to note that reordering only works for symbols at top level scope. Therefore, the following will fail to compile:

{.experimental: "codeReordering".}

proc a() =
  proc b() =


Compiler Messages

The Nim compiler emits different kinds of messages: hint, warning, and error messages. An error message is emitted if the compiler encounters any static error.


Pragmas are Nim's method to give the compiler additional information / commands without introducing a massive number of new keywords. Pragmas are processed on the fly during semantic checking. Pragmas are enclosed in the special {. and .} curly brackets. Pragmas are also often used as a first implementation to play with a language feature before a nicer syntax to access the feature becomes available.

deprecated pragma

The deprecated pragma is used to mark a symbol as deprecated:

proc p() {.deprecated.}
var x {.deprecated.}: char

This pragma can also take in an optional warning string to relay to developers.

proc thing(x: bool) {.deprecated: "use thong instead".}

noSideEffect pragma

The noSideEffect pragma is used to mark a proc/iterator to have no side effects. This means that the proc/iterator only changes locations that are reachable from its parameters and the return value only depends on the arguments. If none of its parameters have the type var T or ref T or ptr T this means no locations are modified. It is a static error to mark a proc/iterator to have no side effect if the compiler cannot verify this.

As a special semantic rule, the built-in debugEcho pretends to be free of side effects, so that it can be used for debugging routines marked as noSideEffect.

func is syntactic sugar for a proc with no side effects:

func `+` (x, y: int): int

compileTime pragma

The compileTime pragma is used to mark a proc or variable to be used at compile time only. No code will be generated for it. Compile time procs are useful as helpers for macros. Since version 0.12.0 of the language, a proc that uses system.NimNode within its parameter types is implicitly declared compileTime:

proc astHelper(n: NimNode): NimNode =
  result = n

Is the same as:

proc astHelper(n: NimNode): NimNode {.compileTime.} =
  result = n

noReturn pragma

The noreturn pragma is used to mark a proc that never returns.

acyclic pragma

The acyclic pragma can be used for object types to mark them as acyclic even though they seem to be cyclic. This is an optimization for the garbage collector to not consider objects of this type as part of a cycle:

  Node = ref NodeObj
  NodeObj {.acyclic.} = object
    left, right: Node
    data: string

Or if we directly use a ref object:

  Node = ref object {.acyclic.}
    left, right: Node
    data: string

In the example a tree structure is declared with the Node type. Note that the type definition is recursive and the GC has to assume that objects of this type may form a cyclic graph. The acyclic pragma passes the information that this cannot happen to the GC. If the programmer uses the acyclic pragma for data types that are in reality cyclic, the GC may leak memory, but nothing worse happens.

Future directions: The acyclic pragma may become a property of a ref type:

  Node = acyclic ref NodeObj
  NodeObj = object
    left, right: Node
    data: string

final pragma

The final pragma can be used for an object type to specify that it cannot be inherited from. Note that inheritance is only available for objects that inherit from an existing object (via the object of SuperType syntax) or that have been marked as inheritable.

shallow pragma

The shallow pragma affects the semantics of a type: The compiler is allowed to make a shallow copy. This can cause serious semantic issues and break memory safety! However, it can speed up assignments considerably, because the semantics of Nim require deep copying of sequences and strings. This can be expensive, especially if sequences are used to build a tree structure:

  NodeKind = enum nkLeaf, nkInner
  Node {.shallow.} = object
    case kind: NodeKind
    of nkLeaf:
      strVal: string
    of nkInner:
      children: seq[Node]

pure pragma

An object type can be marked with the pure pragma so that its type field which is used for runtime type identification is omitted. This used to be necessary for binary compatibility with other compiled languages.

An enum type can be marked as pure. Then access of its fields always requires full qualification.

asmNoStackFrame pragma

A proc can be marked with the asmNoStackFrame pragma to tell the compiler it should not generate a stack frame for the proc. There are also no exit statements like return result; generated and the generated C function is declared as __declspec(naked) or __attribute__((naked)) (depending on the used C compiler).

Note: This pragma should only be used by procs which consist solely of assembler statements.

error pragma

The error pragma is used to make the compiler output an error message with the given content. Compilation does not necessarily abort after an error though.

The error pragma can also be used to annotate a symbol (like an iterator or proc). The usage of the symbol then triggers a compile-time error. This is especially useful to rule out that some operation is valid due to overloading and type conversions:

## check that underlying int values are compared and not the pointers:
proc `==`(x, y: ptr int): bool {.error.}

fatal pragma

The fatal pragma is used to make the compiler output an error message with the given content. In contrast to the error pragma, compilation is guaranteed to be aborted by this pragma. Example:

when not defined(objc):
  {.fatal: "Compile this program with the objc command!".}

warning pragma

The warning pragma is used to make the compiler output a warning message with the given content. Compilation continues after the warning.

hint pragma

The hint pragma is used to make the compiler output a hint message with the given content. Compilation continues after the hint.

line pragma

The line pragma can be used to affect line information of the annotated statement as seen in stack backtraces:

template myassert*(cond: untyped, msg = "") =
  if not cond:
    # change run-time line information of the 'raise' statement:
    {.line: InstantiationInfo().}:
      raise newException(EAssertionFailed, msg)

If the line pragma is used with a parameter, the parameter needs be a tuple[filename: string, line: int]. If it is used without a parameter, system.InstantiationInfo() is used.

linearScanEnd pragma

The linearScanEnd pragma can be used to tell the compiler how to compile a Nim case statement. Syntactically it has to be used as a statement:

case myInt
of 0:
  echo "most common case"
of 1:
  echo "second most common case"
of 2: echo "unlikely: use branch table"
else: echo "unlikely too: use branch table for ", myInt

In the example, the case branches 0 and 1 are much more common than the other cases. Therefore the generated assembler code should test for these values first, so that the CPU's branch predictor has a good chance to succeed (avoiding an expensive CPU pipeline stall). The other cases might be put into a jump table for O(1) overhead, but at the cost of a (very likely) pipeline stall.

The linearScanEnd pragma should be put into the last branch that should be tested against via linear scanning. If put into the last branch of the whole case statement, the whole case statement uses linear scanning.

computedGoto pragma

The computedGoto pragma can be used to tell the compiler how to compile a Nim case in a while true statement. Syntactically it has to be used as a statement inside the loop:

  MyEnum = enum
    enumA, enumB, enumC, enumD, enumE

proc vm() =
  var instructions: array[0..100, MyEnum]
  instructions[2] = enumC
  instructions[3] = enumD
  instructions[4] = enumA
  instructions[5] = enumD
  instructions[6] = enumC
  instructions[7] = enumA
  instructions[8] = enumB
  instructions[12] = enumE
  var pc = 0
  while true:
    let instr = instructions[pc]
    case instr
    of enumA:
      echo "yeah A"
    of enumC, enumD:
      echo "yeah CD"
    of enumB:
      echo "yeah B"
    of enumE:


As the example shows computedGoto is mostly useful for interpreters. If the underlying backend (C compiler) does not support the computed goto extension the pragma is simply ignored.

unroll pragma

The unroll pragma can be used to tell the compiler that it should unroll a for or while loop for runtime efficiency:

proc searchChar(s: string, c: char): int =
  for i in 0 .. s.high:
    {.unroll: 4.}
    if s[i] == c: return i
  result = -1

In the above example, the search loop is unrolled by a factor 4. The unroll factor can be left out too; the compiler then chooses an appropriate unroll factor.

Note: Currently the compiler recognizes but ignores this pragma.

immediate pragma

The immediate pragma is obsolete. See Typed vs untyped parameters.

compilation option pragmas

The listed pragmas here can be used to override the code generation options for a proc/method/converter.

The implementation currently provides the following possible options (various others may be added later).

pragma allowed values description
checks on|off Turns the code generation for all runtime checks on or off.
boundChecks on|off Turns the code generation for array bound checks on or off.
overflowChecks on|off Turns the code generation for over- or underflow checks on or off.
nilChecks on|off Turns the code generation for nil pointer checks on or off.
assertions on|off Turns the code generation for assertions on or off.
warnings on|off Turns the warning messages of the compiler on or off.
hints on|off Turns the hint messages of the compiler on or off.
optimization none|speed|size Optimize the code for speed or size, or disable optimization.
patterns on|off Turns the term rewriting templates/macros on or off.
callconv cdecl|... Specifies the default calling convention for all procedures (and procedure types) that follow.


{.checks: off, optimization: speed.}
# compile without runtime checks and optimize for speed

push and pop pragmas

The push/pop pragmas are very similar to the option directive, but are used to override the settings temporarily. Example:

{.push checks: off.}
# compile this section without runtime checks as it is
# speed critical
# ... some code ...
{.pop.} # restore old settings

register pragma

The register pragma is for variables only. It declares the variable as register, giving the compiler a hint that the variable should be placed in a hardware register for faster access. C compilers usually ignore this though and for good reasons: Often they do a better job without it anyway.

In highly specific cases (a dispatch loop of a bytecode interpreter for example) it may provide benefits, though.

global pragma

The global pragma can be applied to a variable within a proc to instruct the compiler to store it in a global location and initialize it once at program startup.

proc isHexNumber(s: string): bool =
  var pattern {.global.} = re"[0-9a-fA-F]+"
  result = s.match(pattern)

When used within a generic proc, a separate unique global variable will be created for each instantiation of the proc. The order of initialization of the created global variables within a module is not defined, but all of them will be initialized after any top-level variables in their originating module and before any variable in a module that imports it.

pragma pragma

The pragma pragma can be used to declare user defined pragmas. This is useful because Nim's templates and macros do not affect pragmas. User defined pragmas are in a different module-wide scope than all other symbols. They cannot be imported from a module.


when appType == "lib":
  {.pragma: rtl, exportc, dynlib, cdecl.}
  {.pragma: rtl, importc, dynlib: "client.dll", cdecl.}

proc p*(a, b: int): int {.rtl.} =
  result = a+b

In the example a new pragma named rtl is introduced that either imports a symbol from a dynamic library or exports the symbol for dynamic library generation.

Disabling certain messages

Nim generates some warnings and hints ("line too long") that may annoy the user. A mechanism for disabling certain messages is provided: Each hint and warning message contains a symbol in brackets. This is the message's identifier that can be used to enable or disable it:

{.hint[LineTooLong]: off.} # turn off the hint about too long lines

This is often better than disabling all warnings at once.

used pragma

Nim produces a warning for symbols that are not exported and not used either. The used pragma can be attached to a symbol to suppress this warning. This is particularly useful when the symbol was generated by a macro:

template implementArithOps(T) =
  proc echoAdd(a, b: T) {.used.} =
    echo a + b
  proc echoSub(a, b: T) {.used.} =
    echo a - b

# no warning produced for the unused 'echoSub'
echoAdd 3, 5

experimental pragma

The experimental pragma enables experimental language features. Depending on the concrete feature this means that the feature is either considered too unstable for an otherwise stable release or that the future of the feature is uncertain (it may be removed any time).


{.experimental: "parallel".}

proc useParallel() =
    for i in 0..4:
      echo "echo in parallel"

As a top level statement, the experimental pragma enables a feature for the rest of the module it's enabled in. This is problematic for macro and generic instantiations that cross a module scope. Currently these usages have to be put into a .push/pop environment:

# client.nim
proc useParallel*[T](unused: T) =
  # use a generic T here to show the problem.
  {.push experimental: "parallel".}
    for i in 0..4:
      echo "echo in parallel"
import client

Implementation Specific Pragmas

This section describes additional pragmas that the current Nim implementation supports but which should not be seen as part of the language specification.

Bitsize pragma

The bitsize pragma is for object field members. It declares the field as a bitfield in C/C++.

  mybitfield = object
    flag {.bitsize:1.}: cuint


struct mybitfield {
  unsigned int flag:1;

Volatile pragma

The volatile pragma is for variables only. It declares the variable as volatile, whatever that means in C/C++ (its semantics are not well defined in C/C++).

Note: This pragma will not exist for the LLVM backend.

NoDecl pragma

The noDecl pragma can be applied to almost any symbol (variable, proc, type, etc.) and is sometimes useful for interoperability with C: It tells Nim that it should not generate a declaration for the symbol in the C code. For example:

  EACCES {.importc, noDecl.}: cint # pretend EACCES was a variable, as
                                   # Nim does not know its value

However, the header pragma is often the better alternative.

Note: This will not work for the LLVM backend.

Header pragma

The header pragma is very similar to the noDecl pragma: It can be applied to almost any symbol and specifies that it should not be declared and instead the generated code should contain an #include:

  PFile {.importc: "FILE*", header: "<stdio.h>".} = distinct pointer
    # import C's FILE* type; Nim will treat it as a new pointer type

The header pragma always expects a string constant. The string contant contains the header file: As usual for C, a system header file is enclosed in angle brackets: <>. If no angle brackets are given, Nim encloses the header file in "" in the generated C code.

Note: This will not work for the LLVM backend.

IncompleteStruct pragma

The incompleteStruct pragma tells the compiler to not use the underlying C struct in a sizeof expression:

  DIR* {.importc: "DIR", header: "<dirent.h>",
         pure, incompleteStruct.} = object

Compile pragma

The compile pragma can be used to compile and link a C/C++ source file with the project:

{.compile: "myfile.cpp".}

Note: Nim computes a SHA1 checksum and only recompiles the file if it has changed. You can use the -f command line option to force recompilation of the file.

The link pragma can be used to link an additional file with the project:

{.link: "myfile.o".}

PassC pragma

The passC pragma can be used to pass additional parameters to the C compiler like you would using the commandline switch --passC:

{.passC: "-Wall -Werror".}

Note that you can use gorge from the system module to embed parameters from an external command at compile time:

{.passC: gorge("pkg-config --cflags sdl").}

PassL pragma

The passL pragma can be used to pass additional parameters to the linker like you would using the commandline switch --passL:

{.passL: "-lSDLmain -lSDL".}

Note that you can use gorge from the system module to embed parameters from an external command at compile time:

{.passL: gorge("pkg-config --libs sdl").}

Emit pragma

The emit pragma can be used to directly affect the output of the compiler's code generator. So it makes your code unportable to other code generators/backends. Its usage is highly discouraged! However, it can be extremely useful for interfacing with C++ or Objective C code.


{.emit: """
static int cvariable = 420;

{.push stackTrace:off.}
proc embedsC() =
  var nimVar = 89
  # access Nim symbols within an emit section outside of string literals:
  {.emit: ["""fprintf(stdout, "%d\n", cvariable + (int)""", nimVar, ");"].}


For backwards compatibility, if the argument to the emit statement is a single string literal, Nim symbols can be referred to via backticks. This usage is however deprecated.

For a toplevel emit statement the section where in the generated C/C++ file the code should be emitted can be influenced via the prefixes /*TYPESECTION*/ or /*VARSECTION*/ or /*INCLUDESECTION*/:

{.emit: """/*TYPESECTION*/
struct Vector3 {
  Vector3(): x(5) {}
  Vector3(float x_): x(x_) {}
  float x;

type Vector3 {.importcpp: "Vector3", nodecl} = object
  x: cfloat

proc constructVector3(a: cfloat): Vector3 {.importcpp: "Vector3(@)", nodecl}

ImportCpp pragma

Note: c2nim can parse a large subset of C++ and knows about the importcpp pragma pattern language. It is not necessary to know all the details described here.

Similar to the importc pragma for C, the importcpp pragma can be used to import C++ methods or C++ symbols in general. The generated code then uses the C++ method calling syntax: obj->method(arg). In combination with the header and emit pragmas this allows sloppy interfacing with libraries written in C++:

# Horrible example of how to interface with a C++ engine ... ;-)

{.link: "/usr/lib/libIrrlicht.so".}

{.emit: """
using namespace irr;
using namespace core;
using namespace scene;
using namespace video;
using namespace io;
using namespace gui;

  irr = "<irrlicht/irrlicht.h>"

  IrrlichtDeviceObj {.header: irr,
                      importcpp: "IrrlichtDevice".} = object
  IrrlichtDevice = ptr IrrlichtDeviceObj

proc createDevice(): IrrlichtDevice {.
  header: irr, importcpp: "createDevice(@)".}
proc run(device: IrrlichtDevice): bool {.
  header: irr, importcpp: "#.run(@)".}

The compiler needs to be told to generate C++ (command cpp) for this to work. The conditional symbol cpp is defined when the compiler emits C++ code.


The sloppy interfacing example uses .emit to produce using namespace declarations. It is usually much better to instead refer to the imported name via the namespace::identifier notation:

  IrrlichtDeviceObj {.header: irr,
                      importcpp: "irr::IrrlichtDevice".} = object

Importcpp for enums

When importcpp is applied to an enum type the numerical enum values are annotated with the C++ enum type, like in this example: ((TheCppEnum)(3)). (This turned out to be the simplest way to implement it.)

Importcpp for procs

Note that the importcpp variant for procs uses a somewhat cryptic pattern language for maximum flexibility:

  • A hash # symbol is replaced by the first or next argument.
  • A dot following the hash #. indicates that the call should use C++'s dot or arrow notation.
  • An at symbol @ is replaced by the remaining arguments, separated by commas.

For example:

proc cppMethod(this: CppObj, a, b, c: cint) {.importcpp: "#.CppMethod(@)".}
var x: ptr CppObj
cppMethod(x[], 1, 2, 3)


x->CppMethod(1, 2, 3)

As a special rule to keep backwards compatibility with older versions of the importcpp pragma, if there is no special pattern character (any of # ' @) at all, C++'s dot or arrow notation is assumed, so the above example can also be written as:

proc cppMethod(this: CppObj, a, b, c: cint) {.importcpp: "CppMethod".}

Note that the pattern language naturally also covers C++'s operator overloading capabilities:

proc vectorAddition(a, b: Vec3): Vec3 {.importcpp: "# + #".}
proc dictLookup(a: Dict, k: Key): Value {.importcpp: "#[#]".}
  • An apostrophe ' followed by an integer i in the range 0..9 is replaced by the i'th parameter type. The 0th position is the result type. This can be used to pass types to C++ function templates. Between the ' and the digit an asterisk can be used to get to the base type of the type. (So it "takes away a star" from the type; T* becomes T.) Two stars can be used to get to the element type of the element type etc.

For example:

type Input {.importcpp: "System::Input".} = object
proc getSubsystem*[T](): ptr T {.importcpp: "SystemManager::getSubsystem<'*0>()", nodecl.}

let x: ptr Input = getSubsystem[Input]()


x = SystemManager::getSubsystem<System::Input>()
  • #@ is a special case to support a cnew operation. It is required so that the call expression is inlined directly, without going through a temporary location. This is only required to circumvent a limitation of the current code generator.

For example C++'s new operator can be "imported" like this:

proc cnew*[T](x: T): ptr T {.importcpp: "(new '*0#@)", nodecl.}

# constructor of 'Foo':
proc constructFoo(a, b: cint): Foo {.importcpp: "Foo(@)".}

let x = cnew constructFoo(3, 4)


x = new Foo(3, 4)

However, depending on the use case new Foo can also be wrapped like this instead:

proc newFoo(a, b: cint): ptr Foo {.importcpp: "new Foo(@)".}

let x = newFoo(3, 4)

Wrapping constructors

Sometimes a C++ class has a private copy constructor and so code like Class c = Class(1,2); must not be generated but instead Class c(1,2);. For this purpose the Nim proc that wraps a C++ constructor needs to be annotated with the constructor pragma. This pragma also helps to generate faster C++ code since construction then doesn't invoke the copy constructor:

# a better constructor of 'Foo':
proc constructFoo(a, b: cint): Foo {.importcpp: "Foo(@)", constructor.}

Wrapping destructors

Since Nim generates C++ directly, any destructor is called implicitly by the C++ compiler at the scope exits. This means that often one can get away with not wrapping the destructor at all! However when it needs to be invoked explicitly, it needs to be wrapped. The pattern language provides everything that is required:

proc destroyFoo(this: var Foo) {.importcpp: "#.~Foo()".}

Importcpp for objects

Generic importcpp'ed objects are mapped to C++ templates. This means that you can import C++'s templates rather easily without the need for a pattern language for object types:

  StdMap {.importcpp: "std::map", header: "<map>".} [K, V] = object
proc `[]=`[K, V](this: var StdMap[K, V]; key: K; val: V) {.
  importcpp: "#[#] = #", header: "<map>".}

var x: StdMap[cint, cdouble]
x[6] = 91.4


std::map<int, double> x;
x[6] = 91.4;
  • If more precise control is needed, the apostrophe ' can be used in the supplied pattern to denote the concrete type parameters of the generic type. See the usage of the apostrophe operator in proc patterns for more details.
  VectorIterator {.importcpp: "std::vector<'0>::iterator".} [T] = object

var x: VectorIterator[cint]


std::vector<int>::iterator x;

ImportObjC pragma

Similar to the importc pragma for C, the importobjc pragma can be used to import Objective C methods. The generated code then uses the Objective C method calling syntax: [obj method param1: arg]. In addition with the header and emit pragmas this allows sloppy interfacing with libraries written in Objective C:

# horrible example of how to interface with GNUStep ...

{.passL: "-lobjc".}
{.emit: """
#include <objc/Object.h>
@interface Greeter:Object

- (void)greet:(long)x y:(long)dummy;

#include <stdio.h>
@implementation Greeter

- (void)greet:(long)x y:(long)dummy
  printf("Hello, World!\n");

#include <stdlib.h>

  Id {.importc: "id", header: "<objc/Object.h>", final.} = distinct int

proc newGreeter: Id {.importobjc: "Greeter new", nodecl.}
proc greet(self: Id, x, y: int) {.importobjc: "greet", nodecl.}
proc free(self: Id) {.importobjc: "free", nodecl.}

var g = newGreeter()
g.greet(12, 34)

The compiler needs to be told to generate Objective C (command objc) for this to work. The conditional symbol objc is defined when the compiler emits Objective C code.

CodegenDecl pragma

The codegenDecl pragma can be used to directly influence Nim's code generator. It receives a format string that determines how the variable or proc is declared in the generated code.

For variables $1 in the format string represents the type of the variable and $2 is the name of the variable.

The following Nim code:

  a {.codegenDecl: "$# progmem $#".}: int

will generate this C code:

int progmem a

For procedures $1 is the return type of the procedure, $2 is the name of the procedure and $3 is the parameter list.

The following nim code:

proc myinterrupt() {.codegenDecl: "__interrupt $# $#$#".} =
  echo "realistic interrupt handler"

will generate this code:

__interrupt void myinterrupt()

InjectStmt pragma

The injectStmt pragma can be used to inject a statement before every other statement in the current module. It is only supposed to be used for debugging:

{.injectStmt: gcInvariants().}

# ... complex code here that produces crashes ...

compile time define pragmas

The pragmas listed here can be used to optionally accept values from the -d/--define option at compile time.

The implementation currently provides the following possible options (various others may be added later).

pragma description
intdefine Reads in a build-time define as an integer
strdefine Reads in a build-time define as a string
const FooBar {.intdefine.}: int = 5
echo FooBar
nim c -d:FooBar=42 foobar.c

In the above example, providing the -d flag causes the symbol FooBar to be overwritten at compile time, printing out 42. If the -d:FooBar=42 were to be omitted, the default value of 5 would be used.

Custom annotations

It is possible to define custom typed pragmas. Custom pragmas do not effect code generation directly, but their presence can be detected by macros. Custom pragmas are defined using templates annotated with pragma pragma:

template dbTable(name: string, table_space: string = "") {.pragma.}
template dbKey(name: string = "", primary_key: bool = false) {.pragma.}
template dbForeignKey(t: type) {.pragma.}
template dbIgnore {.pragma.}

Consider stylized example of possible Object Relation Mapping (ORM) implementation:

const tblspace {.strdefine.} = "dev" # switch for dev, test and prod environments

  User {.dbTable("users", tblspace).} = object
    id {.dbKey(primary_key = true).}: int
    name {.dbKey"full_name".}: string
    is_cached {.dbIgnore.}: bool
    age: int
  UserProfile {.dbTable("profiles", tblspace).} = object
    id {.dbKey(primary_key = true).}: int
    user_id {.dbForeignKey: User.}: int
    read_access: bool
    write_access: bool
    admin_acess: bool

In this example custom pragmas are used to describe how Nim objects are mapped to the schema of the relational database. Custom pragmas can have zero or more arguments. In order to pass multiple arguments use one of template call syntaxes. All arguments are typed and follow standard overload resolution rules for templates. Therefore, it is possible to have default values for arguments, pass by name, varargs, etc.

Custom pragmas can be used in all locations where ordinary pragmas can be specified. It is possible to annotate procs, templates, type and variable definitions, statements, etc.

Macros module includes helpers which can be used to simplify custom pragma access hasCustomPragma, getCustomPragmaVal. Please consult macros module documentation for details. These macros are no magic, they don't do anything you cannot do yourself by walking AST object representation.

More examples with custom pragmas:

  • Better serialization/deserialization control:
type MyObj = object
  a {.dontSerialize.}: int
  b {.defaultDeserialize: 5.}: int
  c {.serializationKey: "_c".}: string
  • Adopting type for gui inspector in a game engine:
type MyComponent = object
  position {.editable, animatable.}: Vector3
  alpha {.editRange: [0.0..1.0], animatable.}: float32

Foreign function interface

Nim's FFI (foreign function interface) is extensive and only the parts that scale to other future backends (like the LLVM/JavaScript backends) are documented here.

Importc pragma

The importc pragma provides a means to import a proc or a variable from C. The optional argument is a string containing the C identifier. If the argument is missing, the C name is the Nim identifier exactly as spelled:

proc printf(formatstr: cstring) {.header: "<stdio.h>", importc: "printf", varargs.}

Note that this pragma is somewhat of a misnomer: Other backends do provide the same feature under the same name. Also, if one is interfacing with C++ the ImportCpp pragma and interfacing with Objective-C the ImportObjC pragma can be used.

The string literal passed to importc can be a format string:

proc p(s: cstring) {.importc: "prefix$1".}

In the example the external name of p is set to prefixp. Only $1 is available and a literal dollar sign must be written as $$.

Exportc pragma

The exportc pragma provides a means to export a type, a variable, or a procedure to C. Enums and constants can't be exported. The optional argument is a string containing the C identifier. If the argument is missing, the C name is the Nim identifier exactly as spelled:

proc callme(formatstr: cstring) {.exportc: "callMe", varargs.}

Note that this pragma is somewhat of a misnomer: Other backends do provide the same feature under the same name.

The string literal passed to exportc can be a format string:

proc p(s: string) {.exportc: "prefix$1".} =
  echo s

In the example the external name of p is set to prefixp. Only $1 is available and a literal dollar sign must be written as $$.

Extern pragma

Like exportc or importc, the extern pragma affects name mangling. The string literal passed to extern can be a format string:

proc p(s: string) {.extern: "prefix$1".} =
  echo s

In the example the external name of p is set to prefixp. Only $1 is available and a literal dollar sign must be written as $$.

Bycopy pragma

The bycopy pragma can be applied to an object or tuple type and instructs the compiler to pass the type by value to procs:

  Vector {.bycopy.} = object
    x, y, z: float

Byref pragma

The byref pragma can be applied to an object or tuple type and instructs the compiler to pass the type by reference (hidden pointer) to procs.

Varargs pragma

The varargs pragma can be applied to procedures only (and procedure types). It tells Nim that the proc can take a variable number of parameters after the last specified parameter. Nim string values will be converted to C strings automatically:

proc printf(formatstr: cstring) {.nodecl, varargs.}

printf("hallo %s", "world") # "world" will be passed as C string

Union pragma

The union pragma can be applied to any object type. It means all of the object's fields are overlaid in memory. This produces a union instead of a struct in the generated C/C++ code. The object declaration then must not use inheritance or any GC'ed memory but this is currently not checked.

Future directions: GC'ed memory should be allowed in unions and the GC should scan unions conservatively.

Packed pragma

The packed pragma can be applied to any object type. It ensures that the fields of an object are packed back-to-back in memory. It is useful to store packets or messages from/to network or hardware drivers, and for interoperability with C. Combining packed pragma with inheritance is not defined, and it should not be used with GC'ed memory (ref's).

Future directions: Using GC'ed memory in packed pragma will result in compile-time error. Usage with inheritance should be defined and documented.

Unchecked pragma

The unchecked pragma can be used to mark a named array as unchecked meaning its bounds are not checked. This is often useful to implement customized flexibly sized arrays. Additionally an unchecked array is translated into a C array of undetermined size:

  ArrayPart{.unchecked.} = array[0, int]
  MySeq = object
    len, cap: int
    data: ArrayPart

Produces roughly this C code:

typedef struct {
  NI len;
  NI cap;
  NI data[];
} MySeq;

The base type of the unchecked array may not contain any GC'ed memory but this is currently not checked.

Future directions: GC'ed memory should be allowed in unchecked arrays and there should be an explicit annotation of how the GC is to determine the runtime size of the array.

Dynlib pragma for import

With the dynlib pragma a procedure or a variable can be imported from a dynamic library (.dll files for Windows, lib*.so files for UNIX). The non-optional argument has to be the name of the dynamic library:

proc gtk_image_new(): PGtkWidget
  {.cdecl, dynlib: "libgtk-x11-2.0.so", importc.}

In general, importing a dynamic library does not require any special linker options or linking with import libraries. This also implies that no devel packages need to be installed.

The dynlib import mechanism supports a versioning scheme:

proc Tcl_Eval(interp: pTcl_Interp, script: cstring): int {.cdecl,
  importc, dynlib: "libtcl(|8.5|8.4|8.3).so.(1|0)".}

At runtime the dynamic library is searched for (in this order):


The dynlib pragma supports not only constant strings as argument but also string expressions in general:

import os

proc getDllName: string =
  result = "mylib.dll"
  if existsFile(result): return
  result = "mylib2.dll"
  if existsFile(result): return
  quit("could not load dynamic library")

proc myImport(s: cstring) {.cdecl, importc, dynlib: getDllName().}

Note: Patterns like libtcl(|8.5|8.4).so are only supported in constant strings, because they are precompiled.

Note: Passing variables to the dynlib pragma will fail at runtime because of order of initialization problems.

Note: A dynlib import can be overridden with the --dynlibOverride:name command line option. The Compiler User Guide contains further information.

Dynlib pragma for export

With the dynlib pragma a procedure can also be exported to a dynamic library. The pragma then has no argument and has to be used in conjunction with the exportc pragma:

proc exportme(): int {.cdecl, exportc, dynlib.}

This is only useful if the program is compiled as a dynamic library via the --app:lib command line option. This pragma only has an effect for the code generation on the Windows target, so when this pragma is forgotten and the dynamic library is only tested on Mac and/or Linux, there won't be an error. On Windows this pragma adds __declspec(dllexport) to the function declaration.


To enable thread support the --threads:on command line switch needs to be used. The system module then contains several threading primitives. See the threads and channels modules for the low level thread API. There are also high level parallelism constructs available. See spawn for further details.

Nim's memory model for threads is quite different than that of other common programming languages (C, Pascal, Java): Each thread has its own (garbage collected) heap and sharing of memory is restricted to global variables. This helps to prevent race conditions. GC efficiency is improved quite a lot, because the GC never has to stop other threads and see what they reference. Memory allocation requires no lock at all! This design easily scales to massive multicore processors that are becoming the norm.

Thread pragma

A proc that is executed as a new thread of execution should be marked by the thread pragma for reasons of readability. The compiler checks for violations of the no heap sharing restriction: This restriction implies that it is invalid to construct a data structure that consists of memory allocated from different (thread local) heaps.

A thread proc is passed to createThread or spawn and invoked indirectly; so the thread pragma implies procvar.

GC safety

We call a proc p GC safe when it doesn't access any global variable that contains GC'ed memory (string, seq, ref or a closure) either directly or indirectly through a call to a GC unsafe proc.

The gcsafe annotation can be used to mark a proc to be gcsafe, otherwise this property is inferred by the compiler. Note that noSideEffect implies gcsafe. The only way to create a thread is via spawn or createThread. spawn is usually the preferable method. Either way the invoked proc must not use var parameters nor must any of its parameters contain a ref or closure type. This enforces the no heap sharing restriction.

Routines that are imported from C are always assumed to be gcsafe. To disable the GC-safety checking the --threadAnalysis:off command line switch can be used. This is a temporary workaround to ease the porting effort from old code to the new threading model.

To override the compiler's gcsafety analysis a {.gcsafe.} pragma block can be used:

  someGlobal: string = "some string here"
  perThread {.threadvar.}: string

proc setPerThread() =
    deepCopy(perThread, someGlobal)

Future directions:

  • A shared GC'ed heap might be provided.

Threadvar pragma

A variable can be marked with the threadvar pragma, which makes it a thread-local variable; Additionally, this implies all the effects of the global pragma.

var checkpoints* {.threadvar.}: seq[string]

Due to implementation restrictions thread local variables cannot be initialized within the var section. (Every thread local variable needs to be replicated at thread creation.)

Threads and exceptions

The interaction between threads and exceptions is simple: A handled exception in one thread cannot affect any other thread. However, an unhandled exception in one thread terminates the whole process!

Parallel & Spawn

Nim has two flavors of parallelism:

  1. Structured parallelism via the parallel statement.
  2. Unstructured parallelism via the standalone spawn statement.

Nim has a builtin thread pool that can be used for CPU intensive tasks. For IO intensive tasks the async and await features should be used instead. Both parallel and spawn need the threadpool module to work.

Somewhat confusingly, spawn is also used in the parallel statement with slightly different semantics. spawn always takes a call expression of the form f(a, ...). Let T be f's return type. If T is void then spawn's return type is also void otherwise it is FlowVar[T].

Within a parallel section sometimes the FlowVar[T] is eliminated to T. This happens when T does not contain any GC'ed memory. The compiler can ensure the location in location = spawn f(...) is not read prematurely within a parallel section and so there is no need for the overhead of an indirection via FlowVar[T] to ensure correctness.

Note: Currently exceptions are not propagated between spawn'ed tasks!

Spawn statement

spawn can be used to pass a task to the thread pool:

import threadpool

proc processLine(line: string) =
  discard "do some heavy lifting here"

for x in lines("myinput.txt"):
  spawn processLine(x)

For reasons of type safety and implementation simplicity the expression that spawn takes is restricted:

  • It must be a call expression f(a, ...).
  • f must be gcsafe.
  • f must not have the calling convention closure.
  • f's parameters may not be of type var. This means one has to use raw ptr's for data passing reminding the programmer to be careful.
  • ref parameters are deeply copied which is a subtle semantic change and can cause performance problems but ensures memory safety. This deep copy is performed via system.deepCopy and so can be overridden.
  • For safe data exchange between f and the caller a global TChannel needs to be used. However, since spawn can return a result, often no further communication is required.

spawn executes the passed expression on the thread pool and returns a data flow variable FlowVar[T] that can be read from. The reading with the ^ operator is blocking. However, one can use blockUntilAny to wait on multiple flow variables at the same time:

import threadpool, ...

# wait until 2 out of 3 servers received the update:
proc main =
  var responses = newSeq[FlowVarBase](3)
  for i in 0..2:
    responses[i] = spawn tellServer(Update, "key", "value")
  var index = blockUntilAny(responses)
  assert index >= 0
  discard blockUntilAny(responses)

Data flow variables ensure that no data races are possible. Due to technical limitations not every type T is possible in a data flow variable: T has to be of the type ref, string, seq or of a type that doesn't contain a type that is garbage collected. This restriction is not hard to work-around in practice.

Parallel statement


# Compute PI in an inefficient way
import strutils, math, threadpool
{.experimental: "parallel".}

proc term(k: float): float = 4 * math.pow(-1, k) / (2*k + 1)

proc pi(n: int): float =
  var ch = newSeq[float](n+1)
    for k in 0..ch.high:
      ch[k] = spawn term(float(k))
  for k in 0..ch.high:
    result += ch[k]

echo formatFloat(pi(5000))

The parallel statement is the preferred mechanism to introduce parallelism in a Nim program. A subset of the Nim language is valid within a parallel section. This subset is checked to be free of data races at compile time. A sophisticated disjoint checker ensures that no data races are possible even though shared memory is extensively supported!

The subset is in fact the full language with the following restrictions / changes:

  • spawn within a parallel section has special semantics.
  • Every location of the form a[i] and a[i..j] and dest where dest is part of the pattern dest = spawn f(...) has to be provably disjoint. This is called the disjoint check.
  • Every other complex location loc that is used in a spawned proc (spawn f(loc)) has to be immutable for the duration of the parallel section. This is called the immutability check. Currently it is not specified what exactly "complex location" means. We need to make this an optimization!
  • Every array access has to be provably within bounds. This is called the bounds check.
  • Slices are optimized so that no copy is performed. This optimization is not yet performed for ordinary slices outside of a parallel section.

Guards and locks

Apart from spawn and parallel Nim also provides all the common low level concurrency mechanisms like locks, atomic intrinsics or condition variables.

Nim significantly improves on the safety of these features via additional pragmas:

  1. A guard annotation is introduced to prevent data races.
  2. Every access of a guarded memory location needs to happen in an appropriate locks statement.
  3. Locks and routines can be annotated with lock levels to prevent deadlocks at compile time.

Guards and the locks section

Protecting global variables

Object fields and global variables can be annotated via a guard pragma:

var glock: TLock
var gdata {.guard: glock.}: int

The compiler then ensures that every access of gdata is within a locks section:

proc invalid =
  # invalid: unguarded access:
  echo gdata

proc valid =
  # valid access:
  {.locks: [glock].}:
    echo gdata

Top level accesses to gdata are always allowed so that it can be initialized conveniently. It is assumed (but not enforced) that every top level statement is executed before any concurrent action happens.

The locks section deliberately looks ugly because it has no runtime semantics and should not be used directly! It should only be used in templates that also implement some form of locking at runtime:

template lock(a: TLock; body: untyped) =
  {.locks: [a].}:

The guard does not need to be of any particular type. It is flexible enough to model low level lockfree mechanisms:

var dummyLock {.compileTime.}: int
var atomicCounter {.guard: dummyLock.}: int

template atomicRead(x): untyped =
  {.locks: [dummyLock].}:

echo atomicRead(atomicCounter)

The locks pragma takes a list of lock expressions locks: [a, b, ...] in order to support multi lock statements. Why these are essential is explained in the lock levels section.

Protecting general locations

The guard annotation can also be used to protect fields within an object. The guard then needs to be another field within the same object or a global variable.

Since objects can reside on the heap or on the stack this greatly enhances the expressivity of the language:

  ProtectedCounter = object
    v {.guard: L.}: int
    L: TLock

proc incCounters(counters: var openArray[ProtectedCounter]) =
  for i in 0..counters.high:
    lock counters[i].L:
      inc counters[i].v

The access to field x.v is allowed since its guard x.L is active. After template expansion, this amounts to:

proc incCounters(counters: var openArray[ProtectedCounter]) =
  for i in 0..counters.high:
    {.locks: [counters[i].L].}:
        inc counters[i].v

There is an analysis that checks that counters[i].L is the lock that corresponds to the protected location counters[i].v. This analysis is called path analysis because it deals with paths to locations like obj.field[i].fieldB[j].

The path analysis is currently unsound, but that doesn't make it useless. Two paths are considered equivalent if they are syntactically the same.

This means the following compiles (for now) even though it really should not:

{.locks: [a[i].L].}:
  inc i
  access a[i].v

Lock levels

Lock levels are used to enforce a global locking order in order to prevent deadlocks at compile-time. A lock level is an constant integer in the range 0..1_000. Lock level 0 means that no lock is acquired at all.

If a section of code holds a lock of level M than it can also acquire any lock of level N < M. Another lock of level M cannot be acquired. Locks of the same level can only be acquired at the same time within a single locks section:

var a, b: TLock[2]
var x: TLock[1]
# invalid locking order: TLock[1] cannot be acquired before TLock[2]:
{.locks: [x].}:
  {.locks: [a].}:
# valid locking order: TLock[2] acquired before TLock[1]:
{.locks: [a].}:
  {.locks: [x].}:

# invalid locking order: TLock[2] acquired before TLock[2]:
{.locks: [a].}:
  {.locks: [b].}:

# valid locking order, locks of the same level acquired at the same time:
{.locks: [a, b].}:

Here is how a typical multilock statement can be implemented in Nim. Note how the runtime check is required to ensure a global ordering for two locks a and b of the same lock level:

template multilock(a, b: ptr TLock; body: untyped) =
  if cast[ByteAddress](a) < cast[ByteAddress](b):
  {.locks: [a, b].}:

Whole routines can also be annotated with a locks pragma that takes a lock level. This then means that the routine may acquire locks of up to this level. This is essential so that procs can be called within a locks section:

proc p() {.locks: 3.} = discard

var a: TLock[4]
{.locks: [a].}:
  # p's locklevel (3) is strictly less than a's (4) so the call is allowed:

As usual locks is an inferred effect and there is a subtype relation: proc () {.locks: N.} is a subtype of proc () {.locks: M.} iff (M <= N).

The locks pragma can also take the special value "unknown". This is useful in the context of dynamic method dispatching. In the following example, the compiler can infer a lock level of 0 for the base case. However, one of the overloaded methods calls a procvar which is potentially locking. Thus, the lock level of calling g.testMethod cannot be inferred statically, leading to compiler warnings. By using {.locks: "unknown".}, the base method can be marked explicitly as having unknown lock level as well:

type SomeBase* = ref object of RootObj
type SomeDerived* = ref object of SomeBase
  memberProc*: proc ()

method testMethod(g: SomeBase) {.base, locks: "unknown".} = discard
method testMethod(g: SomeDerived) =
  if g.memberProc != nil:

Taint mode

The Nim compiler and most parts of the standard library support a taint mode. Input strings are declared with the TaintedString string type declared in the system module.

If the taint mode is turned on (via the --taintMode:on command line option) it is a distinct string type which helps to detect input validation errors:

echo "your name: "
var name: TaintedString = stdin.readline
# it is safe here to output the name without any input validation, so
# we simply convert `name` to string to make the compiler happy:
echo "hi, ", name.string

If the taint mode is turned off, TaintedString is simply an alias for string.

© 2006–2018 Andreas Rumpf
Licensed under the MIT License.