Constructs

Constructs

Article Type: How-To Audience: Developers, App Admins Module: How To

So far, we have introduced all the parts of the language that allow us to extract data from an input JSON document, combine the data using string and numeric operators, and format the structure of the output JSON document. What follows are the parts that turn this into a Turing-complete, functional programming language.

Comments

JSONata expressions can be interleaved with comments using 'C' style comment delimiters. For example:

/* Long-winded expressions might need some explanation */
(
  $pi := 3.1415926535897932384626;
  /* JSONata is not known for its graphics support! */
  $plot := function($x) {(
    $floor := $string ~> $substringBefore(?, '.') ~> $number;
    $index := $floor(($x + 1) * 20 + 0.5);
    $join([0..$index].('.')) & 'O' & $join([$index..40].('.'))
  )};

  /* Factorial is the product of the integers 1..n */
  $product := function($a, $b) { $a * $b };
  $factorial := function($n) { $n = 0 ? 1 : $reduce([1..$n], $product) };

  $sin := function($x){ /* define sine in terms of cosine */
    $cos($x - $pi/2)
  };
  $cos := function($x){ /* Derive cosine by expanding Maclaurin series */
    $x > $pi ? $cos($x - 2 * $pi) : $x < -$pi ? $cos($x + 2 * $pi) :
      $sum([0..12].($power(-1, $) * $power($x, 2*$) / $factorial(2*$)))
  };

  [0..24].$sin($*$pi/12).$plot($)
)

Conditional Logic

Ternary operator (? :)

If/then/else constructs can be written using the ternary operator ? :.

predicate ? expr1 : expr2

The expression predicate is evaluated. If its effective Boolean value is true, then expr1 is evaluated and returned, otherwise expr2 is evaluated and returned.

Example

Account.Order.Product.{
  `Product Name`: $.Price > 100 ? "Premium" : "Basic"
}

Elvis / Default operator (?:)

The default (or "elvis") operator is syntactic sugar for a common pattern using the ternary operator. It returns the left-hand side if it has an effective Boolean value of true, otherwise it returns the right-hand side.

expr1 ?: expr2

This is equivalent to expr1 ? expr1 : expr2. The elvis operator is useful for providing fallback values when an expression may evaluate to a value with an effective Boolean value of false, without having to repeat the expression twice as you would with the ternary operator.

Example

Account.Order.Product.{
  `Product Name`: $.'Product Name',
  `Category`: $.Category ?: "Uncategorized"
}

Coalescing operator (??)

The coalescing operator is syntactic sugar for a common pattern using the ternary operator with the $exists function. It returns the left-hand side if it is defined (not undefined), otherwise it returns the right-hand side.

expr1 ?? expr2

This is equivalent to $exists(expr1) ? expr1 : expr2. The coalescing operator is useful for providing fallback values only when the left-hand side is missing or not present (empty sequence), but not for other values with an effective Boolean value of false like 0, false, or ''. It avoids having to evaluate the expression twice and explicitly use the $exists function as you would with the ternary operator.

Example

Account.Order.{
  "OrderID": OrderID,
  "Rating": ($sum(Product.Rating) / $count(Product.Rating)) ?? 0
}

Variables

Any name that starts with a dollar $ is a variable. A variable is a named reference to a value. The value can be one of any type in the language's type system.

Built-in variables

• $ — The variable with no name refers to the context value at any point in the input JSON hierarchy.
• $$ — The root of the input JSON. Only needed if you need to break out of the current context to temporarily navigate down a different path, e.g. for cross-referencing or joining data.
• Native (built-in) functions — see the function library.

Variable binding

Values (of any type in the type system) can be bound to variables:

$var_name := "value"

The stored value can be later referenced using the expression $var_name. The scope of a variable is limited to the 'block' in which it was bound. For example:

Invoice.(
  $p := Product.Price;
  $q := Product.Quantity;
  $p * $q
)

This returns Price multiplied by Quantity for the Product in the Invoice.

Functions

The function is a first-class type, and can be stored in a variable just like any other data type. A library of built-in functions is provided and assigned to variables in the global scope. For example, $uppercase contains a function which, when invoked with a string argument str, returns a string with all the characters in str changed to uppercase.

Invoking a function

A function is invoked by following its reference (or definition) by parentheses containing a comma-delimited sequence of arguments.

Examples

• $uppercase("Hello") returns the string "HELLO".
• $substring("hello world", 0, 5) returns the string "hello".
• $sum([1,2,3]) returns the number 6.

Defining a function

Anonymous (lambda) functions can be defined using the following syntax:

function($l, $w, $h){ $l * $w * $h }

and can be invoked using:

function($l, $w, $h){ $l * $w * $h }(10, 10, 5)

which returns 500. The function can also be assigned to a variable for future use (within the block):

(
  $volume := function($l, $w, $h){ $l * $w * $h };
  $volume(10, 10, 5);
)

Function signatures

Functions can be defined with an optional signature which specifies the parameter types of the function. If supplied, the evaluation engine will validate the arguments passed to the function before it is invoked. A dynamic error is thrown if the argument list does not match the signature.

A function signature is a string of the form <params:return>. params is a sequence of type symbols, each one representing an input argument's type. return is a single type symbol representing the return value type.

Type symbols work as follows.

Simple types:

• b — Boolean
• n — number
• s — string
• l — null

Complex types:

• a — array
• o — object
• f — function

Union types:

• (sao) — string, array or object
• (o) — same as o
• u — equivalent to (bnsl), i.e. Boolean, number, string or null
• j — any JSON type. Equivalent to (bnsloa), i.e. Boolean, number, string, null, object or array, but not function
• x — any type. Equivalent to (bnsloaf)

Parametrised types:

• a<s> — array of strings
• a<x> — array of values of any type

Some examples of signatures of built-in JSONata functions:

• $count has signature <a:n>; it accepts an array and returns a number.
• $append has signature <aa:a>; it accepts two arrays and returns an array.
• $sum has signature <a<n>:n>; it accepts an array of numbers and returns a number.
• $reduce has signature <fa<j>:j>; it accepts a reducer function f and an a<j> (array of JSON objects) and returns a JSON object.

Each type symbol may also have options applied:

• + — one or more arguments of this type. E.g. $zip has signature <a+>; it accepts one array, or two arrays, or three arrays, and so on.
• ? — optional argument. E.g. $join has signature <a<s>s?:s>; it accepts an array of strings and an optional joiner string which defaults to the empty string. It returns a string.
• - — if this argument is missing, use the context value ("focus"). E.g. $length has signature <s-:n>; it can be called as $length(OrderID) (one argument) but equivalently as OrderID.$length().

Recursive functions

Functions that have been assigned to variables can invoke themselves using that variable reference. This allows recursive functions to be defined. For example:

(
  $factorial := function($x){ $x <= 1 ? 1 : $x * $factorial($x-1) };
  $factorial(4)
)

returns 24. Note that it is actually possible to write a recursive function using purely anonymous functions (i.e. nothing gets assigned to variables) using the Y-combinator.

Tail call optimization (tail recursion)

A recursive function adds a new frame to the call stack each time it invokes itself. This can eventually lead to stack exhaustion if the function recurses beyond a certain limit. Consider the classic recursive implementation of the factorial function:

(
  $factorial := function($x) {
    $x <= 1 ? 1 : $x * $factorial($x-1)
  };
  $factorial(170)
)

This function works by pushing the number onto the stack, then, when the stack unwinds, multiplying it by the result of the factorial of the number minus one. Written in this way, the JSONata evaluator has no choice but to use the call stack to store the intermediate results. Given a large enough number, the call stack will overflow.

This is a recognised problem with functional programming, and the solution is to rewrite the function slightly to avoid the need for the stack to store the intermediate result. The following implementation of factorial achieves this:

(
  $factorial := function($x){(
    $iter := function($x, $acc) {
      $x <= 1 ? $acc : $iter($x - 1, $x * $acc)
    };
    $iter($x, 1)
  )};
  $factorial(170)
)

Here, the multiplication is done before the function invokes itself, and the intermediate result is carried in the second parameter $acc (accumulator). The invocation of itself is the last thing that the function does. This is known as a 'tail call', and when the JSONata parser spots this, it internally rewrites the recursion as a simple loop. Thus it can run indefinitely without growing the call stack. Functions written in this way are said to be tail recursive.

Higher order functions

A function, being a first-class data type, can be passed as a parameter to another function, or returned from a function. Functions that process other functions are known as higher order functions. Consider the following example:

(
  $twice := function($f) { function($x){ $f($f($x)) } };
  $add3 := function($y){ $y + 3 };
  $add6 := $twice($add3);
  $add6(7)
)

The function stored in variable $twice is a higher order function. It takes a parameter $f which is a function, and returns a function which takes a parameter $x which, when invoked, applies the function $f twice to $x. $add3 stores a function that adds 3 to its argument. Neither $twice nor $add3 have been invoked yet. $twice is invoked by passing the function $add3 as its argument. This returns a function that applies $add3 twice to its argument. This returned function is not invoked yet, but rather assigned to the variable $add6. Finally, the function in $add6 is invoked with the argument 7, resulting in 3 being added to it twice. It returns 13.

Functions are closures

When a lambda function is defined, the evaluation engine takes a snapshot of the environment and stores it with the function body definition. The environment comprises the context item (i.e. the current value in the location path) together with the current in-scope variable bindings. When the lambda function is later invoked, it is done so in that stored environment rather than the current environment at invocation time. This property is known as lexical scoping and is a fundamental property of closures.

Consider the following example:

Account.(
  $AccName := function() { $.'Account Name' };

  Order[OrderID = 'order104'].Product.{
    'Account': $AccName(),
    'SKU-' & $string(ProductID): $.'Product Name'
  }
)

When the function is created, the context item (referred to by $) is the value of Account. Later, when the function is invoked, the context item has moved down the structure to the value of each Product item. However, the function body is invoked in the environment that was stored when it was defined, so its context item is the value of Account. The expression produces the following result:

{
  "Account": "Firefly",
  "SKU-858383": "Bowler Hat",
  "SKU-345664": "Cloak"
}

Partial function application

Functions can be partially applied by invoking the function with one or more (but not all) arguments replaced by a question mark ? placeholder. The result of this is another function whose arity (number of parameters) is reduced by the number of arguments supplied to the original function. This returned function can be treated like any other newly defined function — e.g. bound to a variable, passed to a higher-order function, and so on. For example, you can create a function to return the first five characters of a string by partially applying the $substring function.

See Also

• Construction
• Jsonata-Tutorial
• Expressions
• Composition
• Custom-Fuuz-Only-JSONata-Library

---

Source: support.fuuz.com