Liquidsoap’s scripting language

The following is adapted from the Liquidsoap book. The reader is avised to check out the whole chapter in the book for more details about the liquidsoap language

General features

Liquidsoap is a novel language which was designed from scratch to handle media stream. It takes some inspiration from functional languages such as OCaml but features a syntax that is more intuitive to the general purpose programmer, similar to Ruby or Javascript.


One of the main features of the language is that it is typed. This means that every expression belongs to some type which indicates what it is. For instance, "hello" is a string whereas 23 is an integer, and, when presenting a construction of the language, we will always indicate the associated type. Liquidsoap implements a typechecking algorithm which ensures that whenever a string is expected a string will actually be given, and similarly for other types. This is done without running the program, so that it does not depend on some dynamic tests, but is rather enforced by theoretical considerations. Another distinguishing feature of this algorithm is that it also performs type inference: you never actually have to write a type, those are guessed automatically by Liquidsoap. This makes the language very safe, while remaining very easy to use.

Functional programming

The language is functional, which means that you can very easily define functions, and that functions can be passed as arguments of other functions. This might look like a crazy thing at first, but it is actually quite common in some language communities (such as OCaml). It also might look quite useless: why should we need such functions when describing webradios? You will soon discover that it happens to be quite convenient in many places: for handlers (we can specify the function which describes what to do when some event occurs such as when a DJ connects to the radio), for transitions (we pass a function which describes the shape we want for the transition) and so on.


The unique feature of Liquidsoap is that it allows the manipulation of sources which are functions which will generate streams. These streams typically consist of stereo audio data, but we do restrict to this: they can contain audio with arbitrary number of channels, they can also contain an arbitrary number of video channels, and also MIDI channels (there is limited support for sound synthesis).

Standard library

Although the core of Liquidsoap is written in OCaml, many of the functions of Liquidsoap are written in the Liquidsoap language itself. Those are defined in the stdlib.liq script, which is loaded by default and includes all the libraries. You should not be frightened to have a look at the standard library, it is often useful to better grasp the language, learn design patterns and tricks, and add functionalities. Its location on your system is indicated in the variable configure.libdir and can be obtained by typing

Basic values

Integers and floats

The integers, such as 3 or 42, are of type int. Depending on the current architecture of the computer on which we are executing the script (32 or 64 bits, the latter being the most common nowadays) they are stored on 31 or 63 bits. The minimal (resp. maximal) representable integer can be obtained as the constant min_int (resp. max_int); typically, on a 64 bits architecture, they range from -4611686018427387904 to 4611686018427387903.

The floating point numbers, such as 2.45, are of type float, and are in double precision, meaning that they are always stored on 64 bits. We always write a decimal point in them, so that 3 and 3. are not the same thing: the former is an integer and the latter is a float. This is a source of errors for beginners, but is necessary for typing to work well.


Strings are written between double or single quotes, e.g. "hello!" or 'hello!', and are of type string.

In order to write the character “"” in a string, one cannot simply type “"” since this is already used to indicate the boundaries of a string: this character should be escaped, which means that the character “\” should be typed first so that

print("My name is \"Sam\"!")

will actually display “My name is "Sam"!”. Other commonly used escaped characters are “\\” for backslash and “\n” for new line. Alternatively, one can use the single quote notation, so that previous example can also be written as

print('My name is "Sam"!')

This is most often used when testing JSON data which can contain many quotes or for command line arguments when calling external scripts. The character “\” can also be used at the end of the string to break long strings in scripts without actually inserting newlines in the strings. For instance, the script

print("His name is \

will actually print

His name is Romain.

Note that there is no line change between “is” and “Romain”, and the indentation before “Romain” is not shown either.

The concatenation of two strings is achieved by the infix operator “^”, as in

user = "dj"
print("Current user is " ^ user)

Instead of using concatenation, it is often rather convenient to use string interpolation: in a string, #{e} is replaced by the string representation of the result of the evaluation of the expression e:

user = "admin"
print("The user #{user} has just logged.")

will print The user admin has just logged. or

print("The number #{random.float()} is random.")

will print The number 0.663455738438 is random. (at least it did last time I tried).

Escaping strings

Liquidsoap strings follow the most common lexical conventions from C and javascript and JSON, in particular, string.unescape recognizes the same escape sequences as C (except for UTF-16 characters) and javascript.

The following sequences are recognized:

Escape sequence Hex value in ASCII Character represented
\a \x07 Alert (Beep, Bell)
\b \x08 Backspace
\e \x1B Escape character
\f \x0C Formfeed, Page Break
\n \x0A Newline (Line Feed)
\r \x0D Carriage Return
\t \x09 Horizontal Tab
\v \x0B Vertical Tab
\\ \x5C Backslash
\/ \x2f Forward slash
\' \x27 Apostrophe or single quotation mark
\" \x22 Double quotation mark
\? \x3F Question mark (used to avoid Digraphs and trigraphs)
\nnn any The byte whose numerical value is given by _nnn_interpreted as an octal number
\xhh any The byte whose numerical value is given by hh interpreted as a hexadecimal number
\uhhhh none UTF8-8 code point given by hhhh interpreted as an hexadecimal number

This convention has been decided to follow the most common practices. In particular, \nnn is an octal escape sequence in most languages including C, Ruby, Javascript, Python and more. This differs from OCaml where \nnn is considered a digital escape sequence.

These lexical conventions are used in the default string.escape and string.unescape.

Here’s an example of an escaped string:

# "\" \t \045 \x2f \u4f32";;
- : string = "\" \t % / 2"

The function string.quote returns JSON-compatible strings.


The booleans are either true or false and are of type bool. They can be combined using the usual boolean operations

  • and: conjunction,
  • or: disjunction,
  • not: negation.

Booleans typically originate from comparison operators, which take two values and return booleans:

  • ==: compares for equality,
  • !=: compares for inequality,
  • <=: compares for inequality,

and so on (<, >=, >). For instance, the following is a boolean expression:

(n < 3) and not (s == "hello")

Conditional branchings execute code depending on whether a condition is true or not. For instance, the code

if (1 <= x and x <= 12) or (not 10h-15h) then
  print("The condition is satisfied")
  print("The condition ain't satisified")

will print that the condition is satisfied when either x is between 1 and 12 or the current time is not between 10h and 15h. A conditional branching might return a value, which is the last computed value in the chosen branch. For instance,

y = if x < 3 then "A" else "B" end

will assign "A" or "B" to y depending on whether x is below 3 or not. The two branches of a conditional should always have the same return type:

x = if 1 == 2 then "A" else 5 end

will result in

At line 1, char 19-21:
Error 5: this value has type (...) -> string
but it should be a subtype of (...) -> int

meaning that "A" is a string but is expected to be an integer because the second branch returns an integer, and the two should be of same nature. The else branch is optional, in which case the then branch should be of type unit:

if x == "admin" then print("Welcome admin") end

In the case where you want to perform a conditional branching in the else branch, the elsif keyword should be used, as in the following example, which assigns 0, 1, 2 or 3 to s depending on whether x is "a", "b", "c" or something else:

s = if x == "a" then 0
    elsif x == "b" then 1
    elsif x == "c" then 2
    else 3 end

This is equivalent (but shorter to write) to the following sequence of imbricated conditional branchings:

s = if x == "a" then 0
      if x == "b" then 1
        if x == "c" then 2
        else 3 end

Finally, we should mention that the notation c?a:b is also available as a shorthand for if c then a else b end, so that the expression

y = if x < 3 then "A" else "B" end

can be shortened to

y = (x<3)?"A":"B"

(and people will think that you are a cool guy).

Time predicates

Time predicates are special boolean values such as {0h-7h}. These values are true or false depending on the current time. Some examples of time predicates are

{11h15-13h} between 11h15 and 13h
{12h} between 12h00 and 12h59
{12h00} at 12h00
{00m} on the first minute of every hour
{00m-09m} on the first 10 minutes of every hour
{2w} on Tuesday
{6w-7w} on weekends

Above, w stands for weekday: 1 is Monday, 2 is Tuesday, and so on. Sunday is both 0 and 7.


Some functions, such as print, do not return a meaningful value: we are interested in what they are doing (e.g. printing on the standard output) and not in their result. However, since typing requires that everything returns something of some type, there is a particular type for the return of such functions: unit. Just as there are only two values in the booleans (true and false), there is only one value in the unit type, which is written (). This value can be thought of as the result of the expression saying “I’m done”.


Some more elaborate values can be constructed by combining the previous ones. A first kind is lists which are finite sequences of values, being all of the same type. They are constructed by square bracketing the sequence whose elements are separated by commas. For instance, the list

[1, 4, 5]

is a list of three integers (1, 4 and 5), and its type is [int], and the type of ["A", "B"] would obviously be [string]. Note that a list can be empty: [].

You can extract list elements through splats such as

l = [1, 5, 7, 8, 9]
let [x, _, z, ...t] = l

In this example, the value of x is 1, the value of z is 7 and the value of t is [8, 9].

You can also combine lists in a similar way

x = [1, ...[2, 3, 4], 5, ...[6, 7]]

In this example, the value of x is [1, 2, 3, 4, 5, 6 ,7]


Another construction present in Liquidsoap is tuples of values, which are finite sequences of values which, contrarily to lists, might have different types. For instance,

(3, 4.2, "hello")

is a triple (a tuple with three elements) of type

int * float * string

which indicate that the first element is an integer, the second a float and the third a string.

Similarly to lists, there is a special syntax in order to access tuple elements. For instance, if t is the above tuple (3, 4.2, "hello"), we can write

let (n, x, s) = t

which will assign the first element to the variable n, the second element to the variable x and the third element to the variable s.

Programming primitives


We have already seen many examples of uses of variables: we use

x = e

in order to assign the result of evaluating an expression e to a variable x, which can later on be referred to as x. Variables can be masked: we can define two variables with the same name, and at any point in the program the last defined value for the variable is used:

n = 3
n = n + 2

will print 3 and 5. Contrarily to most languages, the value for a variable cannot be changed (unless we explicitly require this by using references, see below), so the above program does not modify the value of n, it is simply that a new n is defined.

There is an alternative syntax for declaring variables which is

def x =

It has the advantage that the expression e can spread over multiple lines and thus consist of multiple expressions, in which case the value of the last one will be assigned to x. This is particularly useful to use local variables when defining a value.


As indicated above, by default, the value of a variable cannot be changed. However, one can use a reference in order to be able to do this. Those can be seen as memory cells, containing values of a given fixed type, which can be modified during the execution of the program. They are created with the ref keyword, with the initial value of the cell as argument. For instance,

r = ref(5)

declares that r is a reference which contains 5 as initial value. Since 5 is an integer (of type int), the type of the reference r will be


meaning that its a memory cell containing integers. On such a reference, two operations are available:

  • one can obtain the value of the reference by using the ! keyword before the reference, so that !r denotes the value contained in the reference r, for instance

    x = !r + 4

    declares the variable x as being 9 (which is 5+4),

  • one can change the value of the reference by using the := keyword, e.g.

    r := 2

    will assign the value 2 to r.


The usual looping constructions are available in Liquidsoap. The for loop repeatedly executes a portion of code with an integer variable varying between two bounds, being increased by one each time. For instance, the following code will print the integers 1, 2, 3, 4 and 5, which are the values successively taken by the variable i:

for i = 1 to 5 do

In practice, such loops could be used to add a bunch of numbered files (e.g. music1.mp3, music2.mp3, music3.mp3, etc.) in a request queue for instance.

The while loop repeatedly executes a portion of code, as long a condition is satisfied. For instance, the following code doubles the contents of the reference n as long as its value is below 10:

n = ref(1)
while !n < 10 do
  n := !n * 2

The variable n will thus successively take the values 1, 2, 4, 8 and 16, at which point the looping condition !n < 10 is not satisfied anymore and the loop is exited. The printed value is thus 16.


Liquidsoap is built around the notion of function: most operations are performed by those. For some reason, we sometimes call operators the functions acting on sources. Liquidsoap includes a standard library which consists of functions defined in the Liquidsoap language, including fairly complex operators such as playlist which plays a playlist or crossfade which takes care of fading between songs.


A function is a construction which takes a bunch of arguments and produces a result. For instance, we can define a function f taking two float arguments, prints the first and returns the result of adding twice the first to the second:

def f(x, y)

This function can also be written on one line if we use semicolons (;) to separate the instructions instead of changing line:

def f(x, y) = print(x); 2*x+y end

The type of this function is

(int, int) -> int

The arrow -> means that it is a function, on the left are the types of the arguments (here, two arguments of type int) and on the right is the type of the returned value of the function (here, int). In order to use this function, we have to apply it to arguments, as in

f(3, 4)

This will trigger the evaluation of the function, where the argument x (resp. y) is replaced by 3 (resp. 4), i.e., it will print 3 and return the evaluation of 2*3+4, which is 10.

Anonymous functions

For concision in scripts, it is possible define a function without giving it a name, using the syntax

fun (x) -> ...

This is called an anonymous function, and it is typically used in order to specify short handlers in arguments.

Anonymous function with no arguments

You will see that it is quite common to use anonymous functions with no arguments. For this reason, we have introduced a special convenient syntax for those and allow writing


instead of

fun () -> ...

Labeled arguments

A function can have an arbitrary number of arguments, and when there are many of them it becomes difficult to keep track of their order and their order matter! For instance, the following function computes the sample rate given a number of samples in a given period of time:

def samplerate(samples, duration) = samples / duration end

which is of type

(float, float) -> float

For instance, if you have 110250 samples over 2.5 seconds the samplerate will be samplerate(110250., 2.5) which is 44100. However, if you mix the order of the arguments and type samplerate(2.5, 110250.), you will get quite a different result and this will not be detected by the typing system because both arguments have the same type. Fortunately, we can give labels to arguments in order to prevent this, which forces explicitly naming the arguments. This is indicated by prefixing the arguments with a tilde “~”:

def samplerate(~samples, ~duration) = samples / duration end

The labels will be indicated as follows in the type:

(samples : float, duration : float) -> float

Namely, in the above type, we read that the argument labeled samples is a float and similarly for the one labeled duration. For those arguments, we have to give the name of the argument when calling the function:

samplerate(samples=110250., duration=2.5)

The nice byproduct is that the order of the arguments does not matter anymore, the following will give the same result:

samplerate(duration=2.5, samples=110250.)

Of course, a function can have both labeled and non-labeled arguments.

Optional arguments

Another useful feature is that we can give default values to arguments, which thus become optional: if, when calling the function, a value is not specified for such arguments, the default value will be used. For instance, if for some reason we tend to generally measure samples over a period of 2.5 seconds, we can make this become the value for the duration parameter:

In this way, if we do not specify a value for the duration, its value will implicitly be assumed to be 2.5, so that the expression:


will still evaluate to 44100. Of course, if we want to use another value for the duration, we can still specify it, in which case the default value will be ignored:

samplerate(samples=132300., duration=3.)

The presence of an optional argument is indicated in the type by prefixing the corresponding label with “?”, so that the type of the above function is

(samples : float, ?duration : float) -> float


We often want to be able to dynamically modify some parameters in a script. For instance, consider the operator amplify, which takes a float and an audio source and returns the audio amplified by the given volume factor: we can expect its type to be

(float, source('a)) -> source('a)

so that we can use it to have a radio consisting of a microphone input amplified by a factor 1.2 by

mic   = input.alsa()
radio = amplify(1.2, mic)

In the above example, the volume 1.2 was supposedly chosen because the sound delivered by the microphone is not loud enough, but this loudness can vary from time to time, depending on the speaker for instance, and we would like to be able to dynamically update it. The problem with the current operator is that the volume is of type float and a float cannot change over time: it has a fixed value.

In order for the volume to have the possibility to vary over time, instead of having a float argument for amplify, we have decided to have instead an argument of type

() -> float

This is a function which takes no argument and returns a float (remember that a function can take an arbitrary number of arguments, which includes zero arguments). It is very close to a float excepting that each time it is called the returned value can change: we now have the possibility of having something like a float which varies over time. We like to call such a function a float getter, since it can be seen as some kind of object on which the only operation we can perform is get the value. For instance, we can define a float getter by

n = ref(0.)
def f ()
  n := !n + 1.

Each time we call f, by writing f() in our script, the resulting float will be increased by one compared to the previous one: if we try it in an interactive session, we obtain

# f();;
- : float = 1.0
# f();;
- : float = 2.0
# f();;
- : float = 3.0

Since defining such arguments often involves expressions of the form

fun () -> e

which is somewhat heavy, we have introduced the alternative syntax


for it.

Finally, in order to simplify things a bit, you will see that the type of amplify is actually

({float}, source('a)) -> source('a)

where the type {float} means that both float and () -> float are accepted, so that you can still write constant floats where float getters are expected. What we actually call a getter is generally an element of such a type, which is either a constant or a function with no argument.

In order to work with such types, the standard library often uses the following functions:

  • getter, of type ({'a}) -> {'a}, creates a getter,
  • getter.get, of type ({'a}) -> 'a, retrieves the current value of a getter,
  • getter.function, of type ({'a}) -> () -> 'a, creates a function from a getter.

Recursive functions

Liquidsoap supports functions which are recursive, i.e., that can call themselves. For instance, in mathematics, the factorial function on natural numbers is defined as fact(n)=1×2×3×…×n, but it can also be defined recursively as the function such that fact(0)=1 and fact(n)=n×fact(n-1) when n>0: you can easily check by hand that the two functions agree on small values of n (and prove that they agree on all values of n). This last formulation has the advantage of immediately translating to the following implementation of factorial:

def rec fact(n) =
  if n == 0 then 1
  else n * fact(n-1) end

for which you can check that fact(5) gives 120, the expected result. As another example, the list.length function, which computes the length of a list, can be programmed in the following way in Liquidsoap:

def rec length(l)
  if l == [] then 0
  else 1 + length( end

We do not detail much further this trait since it is unlikely to be used for radios, but you can see a few occurrences of it in the standard library.

Partial evaluation

The final thing to know about functions in Liquidsoap is that they support partial evaluation of functions. This means that if you call a function, but do not provide all the arguments, it will return a new function expecting only the remaining arguments. For instance, consider the multiplication function

def mul(x, y) = x * y end

which is of type

(float, float) -> float

taking two floats and returning their products. We can then define a function which will compute the double of its input by

double = mul(2.)

which is of type

(float) -> float

Since we have provided only the first argument to mul, the double will define is still a function waiting for a second argument x and returning mul(2., x), as we can see in the interactive mode:

# def mul(x, y) = x * y end;;
mul : (float, float) -> float = <fun>
# double = mul(2.);;
double : (float) -> float = <fun>
# double(5.);;
- : float = 10.0

Records and modules


Suppose that we want to store and manipulate structured data. For instance, a list of songs together with their duration and tempo. One way to store each song is as a tuple of type string * float * float, but there is a risk of confusion between the duration and the length which are both floats, and the situation would of course be worse if there were more fields. In order to overcome this, one can use a record which is basically the same as a tuple, excepting that fields are named. In our case, we can store a song as

song = { filename = "song.mp3", duration = 257., bpm = 132. }

which is a record with three fields respectively named filename, duration and bpm. The type of such a record is

{filename : string, duration : float, bpm : float}

which indicates the fields and their respective type. In order to access a field of a record, we can use the syntax record.field. For instance, we can print the duration with

print("The duration of the song is #{song.duration} seconds")


Records are heavily used in Liquidsoap in order to structure the functions of the standard library. We tend to call module a record with only functions, but this is really the same as a record. For instance, all the functions related to lists are in the list module and functions such as list.hd are fields of this record. For this reason, the def construction allows adding fields in record. For instance, the definition

def list.last(l)
  list.nth(l, list.length(l)-1)

adds, in the module list, a new field named last, which is a function which computes the last element of a list. Another shorter syntax to perform definitions consists in using the let keyword which allows assigning a value to a field, so that the previous example can be rewritten as

let list.last = fun(l) -> list.nth(l, list.length(l)-1)

If you often use the functions of a specific module, the open keyword allows using its fields without having to prefix them by the module name. For instance, in the following example

l = [1,2,3]
open list
x = nth(l, length(l)-1)

the open list directive allows directly using the functions in this module: we can simply write nth and length instead of list.nth and list.length.

Values with fields

A unique feature of the Liquidsoap language is that it allows adding fields to any value. We also call them methods by analogy with object-oriented programming. For instance, we can write

song = "test.mp3".{duration = 123., bpm = 120.}

which defines a string ("test.mp3") with two methods (duration and bpm). This value has type

string.{duration : float, bpm : float}

and behaves like a string, e.g. we can concatenate it with other strings:

print("the song is " ^ song)

but we can also invoke its methods like a record or a module:

print("the duration is #{song.duration}")

The construction def replaces allows changing the main value while keeping the methods unchanged, so that

def replaces song = "newfile.mp3" end

will print

"newfile.mp3".{duration = 123., bpm = 120.}

(note that the string is modified but not the fields duration and bpm).

Advanced values

In this section, we detail some more advanced values than the ones presented in. You are not expected to be understanding those in details for basic uses of Liquidsoap.


In the case where a function does not have a sensible result to return, it can raise an error. Typically, if we try to take the head of the empty list without specifying a default value (with the optional parameter default), an error will be raised. By default, this error will stop the script, which is usually not a desirable behavior. For instance, if you try to run a script containing


the program will exit printing

Error 14: Uncaught runtime error:
type: not_found, message: "no default value for list.hd"

This means that the error named “not_found” was raised, with a message explaining that the function did not have a reasonable default value of the head to provide.

In order to avoid this, one can catch exceptions with the syntax

catch err do

This will execute the instructions code: if an error is raised at some point during this, the code handler is executed, with err being the error. For instance, instead of writing

l = []
x = list.hd(default=0, l)

we could equivalently write

l = []
x =
  catch err do

The name and message associated to an error can respectively be retrieved using the functions error.kind and error.message, e.g. we can write

catch err do
  print("the error #{error.kind(err)} was raised")
  print("the error message is #{error.message(err)}")

Typically, when reading from or writing to a file, errors will be raised when a problem occurs (such as reading from a non-existent file or writing a file in a non-existent directory) and one should always check for those and log the corresponding message:

data = "bla"
  file.write(data=data, "/non/existent/path")
catch err do
  log.important("Could not write to file: #{error.message(err)}")

Specific errors can be catched with the syntax

catch err : l do

where l is a list of error names that we want to handle here.

Errors can be raised from Liquidsoap with the function error.raise, which takes as arguments the error to raise and the error message. For instance:

error.raise(error.not_found, "we could not find your result")

Finally, we should mention that all the errors should be declared in advance with the function error.register, which takes as argument the name of the new error to register:

myerr = error.register("my_error")
error.raise(myerr, "testing my own error")

Nullable values

It is sometimes useful to have a default value for a type. In Liquidsoap, there is a special value for this, which is called null. Given a type t, we write t? for the type of values which can be either of type t or be null: such a value is said to be nullable. For instance, we could redefine the list.hd function in order to return null (instead of raising an error) when the list is empty:

def list.hd(l)
  if l == [] then null() else list.hd(l) end

whose type would be

(['a]) -> 'a?

since it takes as argument a list whose elements are of type 'a and returns a list whose elements are 'a or null. As it can be observed above, the null value is created with null().

In order to use a nullable value, one typically uses the construction x ?? d which is the value x excepting when it is null, in which case it is the default value d. For instance, with the above head function:

x = list.hd(l)
print("the head is " ^ (x ?? "not defined"))

Some other useful functions include

  • null.defined: test whether a value is null or not,
  • null.get: obtain the value of a nullable value supposed to be distinct from null,
  • execute a function or another, depending on whether a value is null or not.

Including other files

It is often useful to split your script over multiple files, either because it has become quite large, or because you want to be able to reuse common functions between different scripts. You can include a file file.liq in a script by writing

%include "file.liq"

which will be evaluated as if you had pasted the contents of the file in place of the command.

For instance, this is useful in order to store passwords out of the main file, in order to avoid risking leaking those when handing the script to some other people. Typically, one would have a file passwords.liq defining the passwords in variables, e.g.

radio_pass = "secretpassword"

and would then use it by including it:

%include "passwords.liq"

radio = ...
output.icecast(%mp3, host="localhost", port=8000,
               password=radio_pass, mount="my-radio.mp3", radio)

so that passwords are not shown in the main script.