Nickel Syntax

Identifiers

Nickel identifiers start with an alphabetic character, followed by zero or more alphanumeric characters, _ (underscores) or ' (single quotes). For example, this-isn't-invalid is a valid identifier.

Simple values

There are four basic kinds of values in Nickel :

  1. numeric values
  2. boolean values
  3. strings
  4. enum tags

Numeric values

Nickel has support for numbers, positive and negative, with or without decimals. Internally, those numbers are stored as arbitrary precision rationals, meaning that basic arithmetic operations (addition, subtraction, division and multiplication) don't incur rounding errors. Numbers are deserialized as 64-bit floating point numbers, in line with common JSON implementations.

Exponentiation is supported using the std.number.pow function. If the exponent is exactly representable as an integer between -2^63 and 2^64 - 1, the result is computed exactly. However, raising a number to a non-integral power can incur rounding errors: both operands will be converted to the nearest 64-bit floating point numbers, the power is computed as a 64-bit floating point number as well, and then converted back to an arbitrary precision rational number.

Numbers are serialized as integers whenever possible, that is, when they fit exactly into a 64-bit signed integer or a 64-bit unsigned integer. They are serialized as a 64-bit float otherwise. The latter conversion might lose precision as well, for example when serializing 1/3.

Here are some examples of number literals in Nickel; scientific notation for decimal is supported:

1
0.543
1.7e217
-3e-3
-1000000
-6.8

There are some predefined operators for working with numbers:

Operator Description Example
+ The addition operator 1 + 2 = 3
- The subtraction operator 1 - 2 = -1
* The multiplication operator 1 * 2 = 2
/ The division operator 1 / 2 = 0.5
% The modulo operator (returns the signed remainder) 5 % 3 = 2

Remark about the - operator: Since - can be used inside an identifier, the subtraction operators needs to be surrounded by spaces: write a - b, not a-b. 1-2 works as expected, because 1 and 2 aren't identifiers.

Numbers can be compared using the following operators:

Operator Description Example
== Equal 5 == 5
!= Not Equal 5 != 4
< Smaller than 2 < 3
> Greater than 1 > -5
>= Greater or Equal 1 >= 1
<= Smaller or Equal -1 <= 6

In the table below, you will find the operators sorted from highest to lowest precedence:

Operators Associativity Remark
( ... ) parentheses always have the highest precedence
- unary negation (as in -1)
*, /, % left-to-right
+, - left-to-right binary addition and subtraction
<, >, =<, >= left-to-right
==, != left-to-right

Boolean values

The boolean values in Nickel are denoted true and false.

Nickel features the classical boolean operators AND (&&), OR (||) and NOT (!). The AND and OR operators are lazy in the evaluation of the second argument: for example, in exp1 && exp2, exp2 is only evaluated if exp1 evaluates to false.

Here are some examples of boolean operators in Nickel:

> true && false
false

> false || true
true

> ! true
false

Strings

Nickel can work with sequences of characters, or strings. Strings are enclosed by " ... " for a single line string or by m%" ... "% for a multiline string. They can be concatenated with the operator ++. Strings must be UTF-8 valid. In fact, as far as at all practicable, Nickel treats strings as sequences of Unicode extended grapheme clusters and refuses to break them apart.

The string interpolation syntax is "%{ < expression that evaluates to a string > }".

Here are some examples of string handling in Nickel:

> "Hello, World!"
"Hello, World!"

> m%"Well, if this isn't a multiline string?
  Yes it is, indeed it is"%
"Well, if this isn't a multiline string?\n  Yes it is, indeed it is"

> "Hello" ++ "World"
"HelloWorld"

> let h = "Hello" in "%{h} World"
"Hello World"

> let n = 5 in "The number %{n}."
error: dynamic type error
[...]

> let n = 5 in "The number %{std.string.from_number n}."
"The number 5."

Multiline strings are useful for writing indented lines. The first and last lines are ignored if they are empty or contain only spaces. Indentation that is present on all lines of the string is stripped. This way, multiline strings can be indented for nicer code formatting without producing unwanted whitespaces in the output. For example:

> m%"
    This line has no indentation.
      This line is indented.
        This line is even more indented.
    This line has no more indentation.
  "%
"This line has no indentation.\n  This line is indented.\n    This line is even more indented.\nThis line has no more indentation."

The only special sequence in a multiline string is the string interpolation:

> m%"Multiline\nString?"%
"Multiline\\nString?"

> m%"Multiline%{"\n"}String"%
"Multiline\nString"

A multiline string can be opened and closed with multiple % signs, as long as the number of % signs in the start delimiter equals the number in the closing delimiter. If you want to use string interpolation, you must use the same amount of % signs as in the delimiters. This can be useful for writing a literal "% or %{ sequence in a string without escaping:

> m%%"Hello World"%%
"Hello World"

> m%%%%%"Hello World"%%%%%
"Hello World"

> let w = "World" in m%%"Hello %{w}"%%
"Hello \%{w}"

> let w = "World" in m%%"Hello %%{w}"%%
"Hello World"

Multiline string interpolation is "indentation-aware". This means that you can interpolate a string with indentation and the result will be as expected:

> let log = m%"
  if log:
    print("log:", s)
  "% in m%"
  def concat(str_array, log=false):
    res = []
    for s in str_array:
      %{log}
      res.append(s)
    return res
  "%
"def concat(str_array, log=false):\n  res = []\n  for s in str_array:\n    if log:\n      print(\"log:\", s)\n    res.append(s)\n  return res"

Inside a multiline string, an interpolation sequence immediately preceded by a double quote - that is of the form "%..%{ - is always interpreted as a potential string literal prefix "%..% followed by an interpolation sequence opening delimiter, even if the leading "%..% could also act as a string end delimiter:

> let msg = "Hello, world!" in m%"
    echo "%{msg}"
  "%
"echo \"Hello, world!\""

Strictly speaking, the "% prefix of "%{msg} could be interpreted as the end of the multiline string. However, doing so makes the common case of interpolating a variable enclosed in double quotes annoying. Moreover, when such an ambiguity arises, choosing to eagerly interpret "% as a closing delimiter never produces a meaningful expression. Hence, multiline string ending which is followed by zero or more % and an opening brace { is never interpreted as a string ending.

Symbolic Strings

Some tools targeted by Nickel require manipulating string-like values that are not yet known at the time of evaluation, such as Terraform's computed values. Others, like Nix, perform additional dependency tracking (see Nix string context). In both cases, we have to build and combine string-like values which are more complex than bare strings, but for which using a string syntax would still feel natural.

That is precisely the use-case for symbolic strings:

{
  args = [
      "-c",
      nix-s%"
        %{inputs.gcc}/bin/gcc %{inputs.hello} -o hello
        %{inputs.coreutils}/bin/mkdir -p $out/bin
        %{inputs.coreutils}/bin/cp hello $out/bin/hello
      "%,
  ],
  ..
}

This example is an excerpt of a Nix configuration written in Nickel, emulating Nix string contexts. The region delimited by nix-s%" and "%" on lines 4 to 8 is a symbolic string. The values inputs.gcc, inputs.hello, etc. aren't actually strings, but arbitrary records, because they carry additional context. Yet, they can be interpolated as if they were strings.

The idea behind symbolic strings is to offer a string-like syntax, but without evaluating the expression as a string. Instead, the expression is returned in a symbolic form - in practice, an array of fragments, where each fragment is either a string or an arbitrary value that has been interpolated - and Nickel lets the specific library (Terraform-Nickel, Nix-Nickel, etc.) handle it.

The prefix of a symbolic string is any valid identifier that doesn't start with _, and ends with the suffix -s. Prefixes don't have any meaning for Nickel: they're just a tag used by libraries consuming symbolic strings to distinguish between several types of symbolic strings. Prefixes are also a visual marker for the programmer.

Besides the custom prefix, symbolic strings otherwise follow the same syntactic rules as multiline strings: the prefix is followed by an arbitrary number of % followed by ", and must be closed by " followed by the same number of %.

The technical details don't matter too much in practice. As a user of a library which uses symbolic strings, remember that:

  • a special string with a prefix ending in -s is a symbolic string. The prefix (or prefixes) is defined by the library.
  • it's a special syntax without pre-existing meaning for Nickel. The specific meaning of each kind of symbolic string, and what it's used for exactly, is defined by the library. All in all, symbolic strings simply provide libraries with a way to overload string syntax and interpolation for extended usages.
  • the main operation supported by symbolic strings is interpolation: %{value}. What interpolation means, and which values can be interpolated in a given symbolic string is again defined by each library. Other string functions don't work on symbolic strings (e.g. std.string.length, std.string.characters, and so on), because they might not have any valid meaning. Instead, libraries should export their own string API, if they support additional operations on their symbolic strings.

The following examples show how symbolic strings are desugared:

> mytag-s%"I'm %{"symbolic"} with %{"fragments"}"%
{
  fragments = [ "I'm ", "symbolic", " with ", "fragments" ],
  prefix = 'mytag,
  tag = 'SymbolicString,
}

> let terraform_computed_field = {
    tag = 'TfComputed,
    resource = "foo",
    field = "id",
  }

> tf-s%"id: %{terraform_computed_field}, port: %{5}"%
{
  fragments =
    [
        "id: ",
        { field = "id", resource = "foo", tag = 'TfComputed, },
        ", port: ",
        5
      ],
  prefix = 'tf,
  tag = 'SymbolicString,
}

Enum tags

Enumeration tags are used to express a choice among finitely many alternatives. They are formed by writing a single quote ' followed by any valid identifier or by a quoted string. For example, std.serialize takes an export format as a first argument, which is an enum tag among 'Json, 'Toml or 'Yaml:

> std.serialize 'Json {foo = 1}
"{\n  \"foo\": 1\n}"

> std.serialize 'Toml {foo = 1}
"foo = 1\n"

An enum tag 'foo is serialized as the string "foo":

> std.serialize 'Json {foo = 'bar}
"{\n  \"foo\": \"bar\"\n}"

While it's technically possible to just use strings in place of enum tags, using an enum tag encodes the intent that only a finite number of alternatives can be used for the corresponding value.

Additionally, the typechecker is aware of enums and can for example statically enforce that only valid tags are passed to a function within a typed block. See the manual section on typing for more details.

Equality

Operators == and != are used to compare values. Two values of different types are never equal: that is, == doesn't perform implicit conversions.

Here are some examples of equality comparisons in Nickel:

> 1 == 1
true

> 5 == 5.0
true

> "Hello" == "Hello"
true

> "Hello" != "World"
true

> 5 == "Hello"
false

> true == "true"
false

Composite values

Arrays

An array is a sequence of values. Arrays are delimited by [ and ], and elements are separated with ,.

The following are valid Nickel arrays, for example:

[1, 2, 3]
["Hello", "World"]
[1, true, "true"]
[]

Arrays can be concatenated with the operator @:

> [1] @ [2, 3]
[ 1, 2, 3 ]

Records

Records are key-value storage, or in Nickel terms, field-value storage. They are delimited by { and }, and elements are separated with ,. A field definition is written as field = value. The fields are strings, but can be written without quotes " if they are valid identifiers. Values can be of any type. Elements inside a record are unordered. Two records can be merged together using the operator &. The reader can find more information about merging in the section on merging.

Here are some valid Nickel records:

{}
{a = 3}
{my_id_n5 = "my id number 5", "my id n4" = "my id number 4" }
{"5" = 5, six = 6}

Record fields can be accessed using the . operator :

> { a = 1, b = 5 }.a
1

> { a = 1 }.b
error: missing field `b`
[...]

> { "1" = "one" }."1"
"one"

It is possible to write records of records via piecewise syntax, where we separate fields by dots:

> { a = { b = 1 } }
{ a = { b = 1, }, }

> { a.b = 1 }
{ a = { b = 1, }, }

> { a.b = 1, a.c = 2, b = 3}
{ a = { b = 1, c = 2, }, b = 3, }

When fields are enclosed in double quotes ("), you can use string interpolation to create or access fields:

> let k = "a" in { "%{k}" = 1 }
{ a = 1, }

> let k = "a" in { a = 1 }."%{k}"
1

Constructs

If-Then-Else

This construct allows conditional branching in your code. You can use it like if <bool expr> then <expr> else <expr>.

Here are some valid conditional expressions in Nickel:

> if true then "TRUE :)" else "false :("
"TRUE :)"

> if false then "Not this one" else "This one"
"This one"

> if "forty-two" == 42 then "equal?" else "unequal"
"unequal"

> ["1"] @ (if 42 == "42" then ["3"] else ["2"]) @ ["3"]
[ "1", "2", "3" ]

Let-In

A let binds an expression to a variable. It is used like let <rec?> <ident> = <expr> in <expr>. The rec keyword makes the binding recursive, enabling the use of <ident> within the bound expression <expr>. Currently, only a single variable can be bound per let binding.

Here are some examples of let bindings in Nickel:

> let r = { a = "a", b = "b" } in r.a
"a"

> let inner = { inside = true } in let outer = { outside = inner.inside } in outer.outside
true

> let a = 1 in let b = 2 in a + b
3

> let rec f = fun n => if n == 0 then n else n + f (n - 1) in f 10
55

> let rec fib = fun n => if n <= 2 then 1 else fib (n - 1) + fib (n - 2) in fib 9
34

> let rec repeat = fun n x => if n <= 0 then [] else repeat (n - 1) x @ [x] in
    repeat 3 "foo"
[ "foo", "foo", "foo" ]

Functions

A function is declared using the fun keyword, then arguments separated with spaces, and finally an arrow => to add the body of the function. To call a function, just write the arguments after it separated with spaces. Functions in Nickel are curried: a function taking multiple arguments is actually a function that takes a single argument and returns a function taking the rest of the arguments, and so on.

Here are some examples of function definitions in Nickel:

> (fun a b => a + b) 1 2
3

> let add = fun a b => a + b in add 1 2
3

> let add = fun a b => a + b in
    let add1 = add 1 in
        add1 2
3

All existing infix operators in Nickel can be turned into functions by putting them inside parentheses, for example:

> 1 + 2
3

> (+) 1 2
3

> let increment = fun n => (+) 1 n in
    increment 41
42

> let increment = (+) 1 in
    increment 41
42

> let flatten = std.array.fold_right (@) [] in
    flatten [[1, 2], [3], [4, 5]]
[ 1, 2, 3, 4, 5 ]

Functions may be composed using the pipe operator. The pipe operator allows for a function application f x to be written as x |> f. This operator is left-associative, so x |> f |> g will be interpreted as g (f x). For example:

> "Hello World" |> std.string.split " "
[ "Hello", "World" ]

> "Hello World"
  |> std.string.split " "
  |> std.array.first
"Hello"

> "Hello World"
  |> std.string.split " "
  |> std.array.first
  |> std.string.uppercase
"HELLO"

Annotations

Contract and type annotations help enforce additional properties of an expression. They can be attached to any Nickel expression. See the correctness section for more details.

Type annotations

A type annotation is introduced using <expr> : <type> and serves to delimit a statically typed block which will be checked by the typechecker before evaluation. A type annotation can be directly attached to the variable of a let- binding let <var> : <type> = <expr> in <body> or to a record field declaration {<field> : <type> = <value>} as well.

A type wildcard _ indicates that part of a type is unknown to the user (or is not worth spelling out). The typechecker will attempt to infer it. Adding a wildcard type annotation : _ to an existing expression is particularly useful for debugging, as it's the simplest way to have the typechecker run on an expression without having to come up with a type.

Here are some examples of type annotations in Nickel:

> 5 : Number
5

> "hello" : String
"hello"

> "Hello," ++ " world!" : String
"Hello, world!"

> 5 + "a" : _
error: incompatible types
[...]

> (1 + 1 : Number) + (('foo |> match { 'foo => 1, _ => 2 }) : Number)
3

> let x : Number = "a" in x
error: incompatible types
[...]

> let complex_argument : _ -> Number = fun {field1, field2, field3} => field1 in
    complex_argument {field1 = 5, field2 = null, field3 = false}
5

Contract annotations

A contract annotation is introduced by <exp> | <contract> and serves to apply a runtime check to an expression (among other things).

As detailed in the next section, <expr>, <type> and <contract> are in fact syntactically all the same and can be arbitrary Nickel expressions in practice.

Here are some examples of contract annotations in Nickel:

> 5 | Number
5

> 5 | Bool
error: contract broken by a value
[...]

> let SmallNumber = std.contract.from_predicate (fun x => x < 5) in
  1 | SmallNumber
1

> let SmallNumber = std.contract.from_predicate (fun x => x < 5) in
  10 | SmallNumber
error: contract broken by a value
[...]

> let SmallNumber = std.contract.from_predicate (fun x => x < 5) in
  let NotTooSmallNumber = std.contract.from_predicate (fun x => x >= 2) in
  3 | Number
    | SmallNumber
    | NotTooSmallNumber
3

Types

The Nickel syntax mixes both terms and types in the same namespace. The following program is perfectly legal: let value = Number -> (fun value label => value) in ((fun x => x + 1) : value). See the RFC002 for a detailed account of this design.

The documentation still makes a distinction between types and other expressions, the former being constructs which are handled specially by the typechecker and are listed below. However, any expression can be considered a type (in the generic case, it will be considered as an opaque type), and type constructors can also appear inside an expression (where they are understood as their associated contract, which is indeed an expression, most often a function).

Thus, placeholders such as <source>, <target> or <type> can actually be substituted with any valid Nickel expression (which includes the type constructors we've just listed), and types can appear anywhere.

Nickel features the following builtin types and type constructors:

  • Primitive types: Number, String, Bool, and Dyn (the dynamic type, which

represents any value)

  • Arrays: Array <type> is an array whose elements are of type <type>.
  • Dictionaries: {_ : <type>} is a record whose fields are of type <type>.
  • Enums: [| 'tag1, .., 'tagn |] is an enumeration comprised of the alternatives 'tag1, .., 'tagn. Tags have the same syntax as identifiers and must be prefixed with a single quote '. Like record fields, they can however be enclosed in double quotes if they contain special characters: '"tag with space".
  • Arrows: <source> -> <target> is a function taking an argument of type <source> and returns values of type <target>.
  • Foralls: forall var1 .. varn. <type> is a polymorphic type quantifying over type

variables var1, .., varn.

Type variables bound by a forall are only visible inside types (any of the constructor listed above). As soon as a term expression appears under a forall binder, the type variables aren't in scope anymore:

> forall a. a -> (a -> a) -> {_ : {foo : a}}
<func>

> forall a. a -> (a -> (fun x => a))
error: unbound identifier `a`
[...]

Here are some examples of more complicated types in Nickel:

> let f : forall a. a -> a = fun x => x in (f 5 : Number)
5

> {foo = [[1]]} : {foo : Array (Array Number)}
{ foo = [ [ 1 ] ], }

> let select
    : forall a. {left: a, right: a} -> [| 'left, 'right |] -> a
    = fun {left, right} =>
       match {
         'left => left,
         'right => right,
       }
  in
  (select {left = true, right = false} 'left) : Bool
true

> let add_foo : forall a. {_: a} -> a -> {_: a} = fun dict value =>
    std.record.insert "foo" value dict
  in
  add_foo {bar = 1} 5 : _
{ bar = 1, foo = 5, }

> {foo = 1, bar = "string"} : {_ : Number}
error: incompatible types
  ┌─ <repl-input-77>:1:181{foo = 1, bar = "string"} : {_ : Number}
  │                  ^^^^^^^^ this expression
  │
  = Expected an expression of type `Number`
  = Found an expression of type `String`
  = These types are not compatible

Record types

Record types are syntactically a restricted subset of record literals. They are handled differently than normal record literals with respect to typechecking.

A record literal is a record type if:

  • No field has a defined value: there are only fields without definition.
  • Each field has exactly one type annotation
  • Each field doesn't have any other metadata attached (see Metadata)

If these properties are satisfied, a record literal is considered to be a record type by the typechecker.

A record literal which is interpreted as a record type may have a record tail. A tail is written like { <fields> ; <tail> }. It appears at the end of the field declarations and is preceded by ;. The tail <tail> itself must be a valid identifier.

Trying to attach a tail ; tail to a record literal which isn't a record type is a parse error.

Here are some examples of record types in Nickel:

> {foo = 1, bar = "foo" } : {foo : Number, bar: String}
{ bar = "foo", foo = 1, }

> {foo.bar = 1, baz = 2} : {foo: {bar : Number}, baz : Number}
{ baz = 2, foo = { bar = 1, }, }

Here, the right-hand side is missing a type annotation for baz, so it doesn't qualify as a record type and is parsed as a record contract. This throws an "incompatible types" error:

> {foo = 1, bar = "foo" } : {foo : Number, bar : String, baz : Bool}
error: type error: missing row `baz`
[...]

If there's a metadata annotation apart from the type, the record cannot be parsed as a type. Consequently, Nickel tries to interpret it as a record contract which will most likely result in an error, because fields with a type annotation but no value are forbidden outside of types.

> {foo = 1, bar = "foo" } : {foo : Number, bar : String | optional}
error: statically typed field without a definition
  ┌─ <repl-input-81>:1:291{foo = 1, bar = "foo" } : {foo : Number, bar : String | optional}
  │                             ^^^   ------ but it has a type annotation
  │                             │
  │                             this field doesn't have a definition
  │
  = A static type annotation must be attached to an expression but this field doesn't have a definition.
  = Did you mean to use `|` instead of `:`, for example when defining a record contract?
  = Typed fields without definitions are only allowed inside record types, but the enclosing record literal doesn't qualify as a record type. Please refer to the manual for the defining conditions of a record type.

While in the following MyDyn isn't a proper type, the record literal {foo : Number, bar : MyDyn} respects all the requirements for a record type and is parsed as such:

> let MyDyn = fun label value => value in
    {foo = 1, bar | MyDyn = "foo"} : {foo : Number, bar : MyDyn}
{ bar | MyDyn = "foo", foo = 1, }

Metadata

Metadata annotations are used to attach type and contract annotations, documentation, a merge priority or other decorations to record fields (and record fields only). Multiple metadata annotations can be chained. Metadata is introduced with the syntax <field_name> | <metadata1> | .. | <metadataN> [= value].

Documentation can be attached with | doc <string>. For example:

> let record = {
    value
      | doc "The number five"
      | default
      = 5
  }

> :query record value
• default: 5
• documentation: The number five

> {
    truth
      | Bool
      | doc m%"
          If something is true,
          it is based on facts rather than being invented or imagined,
          and is accurate and reliable.
          (Collins dictionary)
        "%
      = true,
  }.truth
true

Metadata can also set merge priorities using the following annotations:

  • default is the lowest priority, usually used for default values that are expected to be overridden somewhere
  • priority NN, where NN is a number literal, is a numeral priority
  • force is the highest priority

If there is no priority specified, priority 0 is the default. See more about this in the dedicated section on merging.

Here are some examples using merge priorities in Nickel:

> let Ais2ByDefault = { a | default = 2 } in
    {} | Ais2ByDefault
{ a | default = 2, }

> let Ais2ByDefault = { a | default = 2 } in
    { a = 1 } | Ais2ByDefault
{ a = 1, }

> { foo | default = 1, bar = foo + 1 }
{ bar = 2, foo | default = 1, }

> {foo | default = 1, bar = foo + 1} & {foo = 2}
{ bar = 3, foo = 2, }

> {foo | force = 1, bar = foo + 1} & {foo = 2}
{ bar = 2, foo | force = 1, }

> {foo | priority 10 = 1} & {foo | priority 8 = 2} & {foo = 3}
{ foo | priority 10 = 1, }

> {foo | priority -1 = 1} & {foo = 2}
{ foo = 2, }

The optional annotation indicates that a field is not mandatory. It is usually found in record contracts.

> let Contract = {
    foo | Number,
    bar | Number
        | optional,
  }

> let value | Contract = {foo = 1}

> value
{ foo | Number = 1, }

> {bar = 1} | Contract
error: missing definition for `foo`
[...]

The not_exported annotation indicates that a field should be skipped when a record is serialized. This includes the output of the nickel export command:

> let value = { foo = 1, bar | not_exported = 2}

> value
{ bar = 2, foo = 1, }

> std.serialize 'Json value
"{\n  \"foo\": 1\n}"