Importing JSON values

Note: If you are reading this page for the first time, you might want to skip directly to the explicit type annotation below as this is the recommended way of parsing JSON data. The content before that is here to explain the inner workings of JSON parsing in liquidsoap.

Liquidsoap supports importing JSON values through a special let syntax. Using this syntax makes it relatively natural to parse JSON data in your script while keeping type-safety at runtime. Here’s an example:

let json.parse v = '{"foo": "abc"}'

print("We parsed a JSON object and got value " ^ v.foo ^ " for attribute foo!")

This prints:

We parsed a JSON object and got value abc for attribute foo!

What happened here is that liquidsoap kept track of the fact that v was called with v.foo and that the result of that was a string. Then, at runtime, it checks the parsed JSON value against this type and raises an issue if that did not match. For instance, the following script:

let json.parse v = '{"foo": 123}'

print("We parsed a JSON object and got value " ^ v.foo ^ " for attribute foo!")

raises the following exception:

Error 14: Uncaught runtime error:
type: json,
message: "Parsing error: json value cannot be parsed as type {foo: string, _}"

Of course, this all seems pretty trivial presented like that but, let’s switch to reading a file instead:

let json.parse v = file.contents("/path/to/file.json")

print("We parsed a JSON object and got value " ^ v.foo ^ " for attribute foo!")

Now, this is getting somewhere! Let’s push it further and parse a whole package.json from a typical npm package:

# Content of package.json is:
# {
#  "name": "my_package",
#  "version": "1.0.0",
#  "scripts": {
#    "test": "echo \"Error: no test specified\" && exit 1"
#  },
#  ...
let json.parse package = file.contents("/path/to/package.json")

name = package.name
version = package.version
test = package.scripts.test

print("This is package " ^  name ^ ", version " ^ version ^ " with test script: " ^ test)

And we get:

This is package my_package, version 1.0.0 with test script: echo "Error: no test specified" && exit 1

This can even be combined with patterns:

let json.parse {
  name,
  version,
  scripts = {
    test
  }
} = file.contents("/path/to/package.json")

print("This is package " ^  name ^ ", version " ^ version ^ " with test script: " ^ test)

Now, this is looking nice!

Explicit type annotation

Explicit type annotation are the recommended way to parse JSON data.

Let’s try a slight variation of the previous script now:

let json.parse {
  name,
  version,
  scripts = {
    test
  }
} = file.contents("/path/to/package.json")

print("This is package #{name}, version #{version} with test script: #{test}")

This returns:

This is package null, version null with test script: null

What? 🤔

This is because, in this script, we only use name, version, etc.. through the interpolation syntax #{...}. However, interpolated variables can be anything so this does not leave enough information to the typing system to know what type those variables should be and, in this case, we default to null.

In order to avoid bad surprises like this, it is usually recommended to add type annotations to your json parsing call to explicitely state what kind of data you are expecting. Let’s add one here:

let json.parse ({
  name,
  version,
  scripts = {
    test
  }
} : {
  name: string,
  version: string,
  scripts: {
    test: string
  }
}) = file.contents("/path/to/package.json")

print("This is package #{name}, version #{version} with test script: #{test}")

And we get:

This is package my_package, version 1.0.0 with test script: echo "Error: no test specified" && exit 1

Back to normal!

Type syntax

The syntax for type annotation is as follows:

Ground types

string, int, float are parsed as, resp., a string, an integer or a floating point number. Note that if your json value contains an integer such as 123, parsing it as a floating point number will succeed. Also, if an integer is too big to be represented as an int internally, it will be parsed as a floating point number.

Nullable types

All type annotation can be postfixed with a trailing ? to denote a nullable value. If a type is nullable, the json parser will return null when it cannot parse the value as the principal type. This is particularly useful when you are not sure of all the types that you are parsing.

For instance, some npm packages do not have a scripts entry or a test entry, so you would parse them as:

let json.parse ({
  name,
  version,
  scripts,
} : {
  name: string,
  version: string,
  scripts: {
    test: string?
  }?
}) = file.contents("/path/to/package.json")

And, later, inspect the returned value to see if it is in fact present.

Tuple types

The type (int * float * string) tells liquidsoap to parse a JSON array whose first values are of type: int, float and string. If any further values are present in the array, they will be ignored.

For arrays as well as any other structured types, the special notation _ can be used to denote any type. For instance, (_ * _ * float) denotes an JSON array whose first 2 elements can be of any type and its third element is a floating point number.

Lists

The type [int] tells liquidsoap to parse a JSON array where all its values are integers as a list of integers. If you are not sure if all elements in the array are integers, you can always use nullable integers: [int?]

Objects

The type {foo: int} tells liquidsoap to parse a JSON object as a record with an attribute labelled foo whose value is an integer. All other attributes are ignored.

Arbitrary object keys can be parsed using the following syntax: {"foo bar key" as foo_bar_key: int}, which tells liquidsoap to parse a JSON object as a record with an attribute labelled foo_bar_key which maps to the attribute "foo bar key" from the JSON object.

Associative lists as objects

It can sometimes be useful to parse a JSON object as an associative list, for instance if you do not know in advance all the possible keys of an object. In this case, you can use the special type: [(string * int)] as json.object. This tells liquidsoap to parse the JSON object as a list of pairs (string * int) where string represents the attribute label and int represent the attribute value.

If you are not sure if all the object values are integers you can always use nullable integers: [(string * int?)] as json.object

Example

Here’s a full example. Feel free to refer to tests/language/json.liq in the source code for more of them.

  data = '{
    "foo": 34.24,
    "gni gno": true,
    "nested": {
       "tuple": [123, 3.14, false],
       "list":  [44.0, 55, 66.12],
       "nullable_list": [12.33, 23, "aabb"],
       "object_as_list": {
         "foo": 123,
         "gni": 456.0,
         "gno": 3.14
       },
       "arbitrary object key ✨": true
     },
     "extra": "ignored"
  }'

  let json.parse ( x : {
    foo: float,
    "gni gno" as gni_gno: bool,
    nested: {
      tuple: (_ * float),
      list: [float],
      nullable_list: [int?],
      object_as_list: [(string * float)] as json.object,
      "arbitrary object key ✨" as arbitrary_object_key: bool,
      not_present: bool?
    }
  }) = data
  - x : {
    foo = 34.24,
    gni_gno = true,
    nested = {
      tuple = (null, 3.14),
      list = [44., 55., 66.12],
      nullable_list = [null, 23, null],
      object_as_list = [("foo", 123.), ("gni", 456.0), ("gno", 3.14)],
      arbitrary_object_key = true,
      not_present = null
    }
  }

JSON5 extension

Liquidsoap supports the JSON5 extension. Parsing of json5 values is enabled with the following argument:

let json.parse[json5=true] x = ...

If a json5 variable is in scope, you can also simply use let json.parse[json5] x = ...

Exporting JSON values

Exporting JSON values works similarly to importing, using the json.stringify let syntax:

let json.stringify s = x

You can also use type annotation to explicitely state what kind of JSON value you want to render:

let json.stringify s = (x : {foo: int})

The json.stringify syntax supports the following arguments:

  • json5: allow json5 representations, in particular infinite and NaN floating point numbers. Default: false
  • compact: output a compact value instead of a human-readable value. Default: false.

Generic JSON objects

Generic JSON objects can be manipulated through the json() operator. This operator returns an opaque json variable with methods to add and remove attributes:

j = json()
j.add("foo", 1)
j.add("bla", "bar")
j.add("baz", 3.14)
j.add("key_with_methods", "value".{method = 123})
j.add("record", { a = 1, b = "ert"})
j.remove("foo")
let json.stringify s = j
- s: '{ "record": { "b": "ert", "a": 1 }, "key_with_methods": "value", "bla": "bar", "baz": 3.14 }'