dhall-1.33.1: A configuration language guaranteed to terminate

Safe HaskellNone
LanguageHaskell2010

Dhall.Tutorial

Contents

Description

Dhall is a programming language specialized for configuration files. This module contains a tutorial explaining how to author configuration files using this language

Synopsis

    Introduction

    The simplest way to use Dhall is to ignore the programming language features and use it as a strongly typed configuration format. For example, suppose that you create the following configuration file:

    -- ./config.dhall
    { foo = 1
    , bar = [3.0, 4.0, 5.0]
    }

    You can read the above configuration file into Haskell using the following code:

    -- example.hs
    
    {-# LANGUAGE DeriveGeneric     #-}
    {-# LANGUAGE OverloadedStrings #-}
    
    import Dhall
    
    data Example = Example { foo :: Natural, bar :: Vector Double }
        deriving (Generic, Show)
    
    instance FromDhall Example
    
    main :: IO ()
    main = do
        x <- input auto "./config.dhall"
        print (x :: Example)

    WARNING: You must not instantiate FromDhall with a recursive type. See Limitations.

    If you compile and run the above example, the program prints the corresponding Haskell record:

    $ ./example
    Example {foo = 1, bar = [3.0,4.0,5.0]}

    You can also load some types directly into Haskell without having to define a record, like this:

    >>> import Dhall
    >>> :set -XOverloadedStrings
    >>> input auto "True" :: IO Bool
    True
    

    The input function can decode any value if we specify the value's expected Type:

    input
        :: Type a  -- Expected type
        -> Text    -- Dhall program
        -> IO a    -- Decoded expression

    ... and we can either specify an explicit type like bool:

    bool :: Type Bool
    
    input bool :: Text -> IO Bool
    
    input bool "True" :: IO Bool
    >>> input bool "True"
    True
    

    ... or we can use auto to let the compiler infer what type to decode from the expected return type:

    auto :: FromDhall a => Type a
    
    input auto :: FromDhall a => Text -> IO a
    >>> input auto "True" :: IO Bool
    True
    

    You can see what types auto supports "out-of-the-box" by browsing the instances for the FromDhall class. For example, the following instance says that we can directly decode any Dhall expression that evaluates to a Bool into a Haskell Bool:

    instance FromDhall Bool

    ... which is why we could directly decode the string "True" into the value True.

    There is also another instance that says that if we can decode a value of type a, then we can also decode a List of values as a Vector of as:

    instance FromDhall a => FromDhall (Vector a)

    Therefore, since we can decode a Bool, we must also be able to decode a List of Bools, like this:

    >>> input auto "[True, False]" :: IO (Vector Bool)
    [True,False]
    

    We could also specify what type to decode by providing an explicit Type instead of using auto with a type annotation:

    >>> input (vector bool) "[True, False]"
    [True,False]
    

    Exercise: Create a ./config.dhall file that the following program can decode:

    {-# LANGUAGE DeriveAnyClass    #-}
    {-# LANGUAGE DeriveGeneric     #-}
    {-# LANGUAGE OverloadedStrings #-}
    
    import Dhall
    
    data Person = Person { age :: Natural, name :: Text }
        deriving (Generic, FromDhall, Show)
    
    main :: IO ()
    main = do
        x <- input auto "./config.dhall"
        print (x :: Person)

    Exercise: Create a ./config.dhall file that the following program can decode:

    {-# LANGUAGE OverloadedStrings #-}
    
    import Data.Functor.Identity
    import Dhall
    
    instance FromDhall a => FromDhall (Identity a)
    
    main :: IO ()
    main = do
        x <- input auto "./config.dhall"
        print (x :: Identity Double)

    Types

    Suppose that we try to decode a value of the wrong type, like this:

    >>> input auto "1" :: IO Bool
    *** Exception:
    ...Error...: Expression doesn't match annotation
    ...
    - Bool
    + Natural
    ...
    1│ 1 : Bool
    ...
    (input):1:1
    ...
    

    The interpreter complains because the string "1" cannot be decoded into a Haskell value of type Bool. This part of the type error:

    - Bool
    + Natural

    ... means that the expected type was Bool but the inferred type of the expression 1 was Natural.

    The code excerpt from the above error message has two components:

    • the expression being type checked (i.e. 1)
    • the expression's expected type (i.e. Bool)
    Expression
    ⇩
    1 : Bool
        ⇧
        Expected type

    The (:) symbol is how Dhall annotates values with their expected types. This notation is equivalent to type annotations in Haskell using the (::) symbol. Whenever you see:

    x : t

    ... you should read that as "we expect the expression x to have type t". However, we might be wrong and if our expected type does not match the expression's actual type then the type checker will complain.

    In this case, the expression 1 does not have type Bool so type checking fails with an exception.

    Exercise: Load the Dhall library into ghci and run these commands to get get a more detailed error message:

    >>> import Dhall
    >>> :set -XOverloadedStrings
    >>> detailed (input auto "1") :: IO Bool
    ...

    ... then read the entire error message

    Exercise: Fix the type error, either by changing the value to decode or changing the expected type

    Imports

    You might wonder why in some cases we can decode a configuration file:

    >>> writeFile "bool" "True"
    >>> input auto "./bool" :: IO Bool
    True
    

    ... and in other cases we can decode a value directly:

    >>> input auto "True" :: IO Bool
    True
    

    This is because importing a configuration from a file is a special case of a more general language feature: Dhall expressions can reference other expressions by their file path.

    To illustrate this, let's create three files:

    $ echo "True"  > bool1
    $ echo "False" > bool2
    $ echo "./bool1 && ./bool2" > both

    ... and read in all three files in a single expression:

    >>> input auto "[ ./bool1 , ./bool2 , ./both ]" :: IO (Vector Bool)
    [True,False,False]
    

    Each file path is replaced with the Dhall expression contained within that file. If that file contains references to other files then those references are transitively resolved.

    In other words: configuration files can reference other configuration files, either by their relative or absolute paths. This means that we can split a configuration file into multiple files, like this:

    -- ./config.dhall
    { foo = 1
    , bar = ./bar.dhall
    }
    -- ./bar.dhall
    [3.0, 4.0, 5.0]

    ... which is equivalent to our original configuration:

    -- ./config.dhall
    { foo = 1
    , bar = [3.0, 4.0, 5.0]
    }

    However, the Dhall language will forbid cycles in these file references. For example, if we create the following cycle:

    $ echo './file1' > file2
    $ echo './file2' > file1

    ... then the interpreter will reject the import:

    >>> input auto "./file1" :: IO Natural
    *** Exception: 
    ↳ ./file1
      ↳ ./file2
    ...
    Cyclic import: ./file1
    ...
    

    You can also import expressions by URL. For example, you can find a Dhall expression hosted at this GitHub URL:

    https://raw.githubusercontent.com/dhall-lang/dhall-haskell/18e4e9a18dc53271146df3ccf5b4177c3552236b/examples/True

    $ curl https://raw.githubusercontent.com/dhall-lang/dhall-haskell/18e4e9a18dc53271146df3ccf5b4177c3552236b/examples/True
    True

    ... and you can reference that expression either directly:

    >>> input auto "https://raw.githubusercontent.com/dhall-lang/dhall-haskell/18e4e9a18dc53271146df3ccf5b4177c3552236b/examples/True" :: IO Bool
    True
    

    ... or inside of a larger expression:

    >>> input auto "False == https://raw.githubusercontent.com/dhall-lang/dhall-haskell/18e4e9a18dc53271146df3ccf5b4177c3552236b/examples/True" :: IO Bool
    False
    

    You're not limited to hosting Dhall expressions on GitHub. You can host a Dhall expression anywhere that you can host UTF8-encoded text on the web, such as a pastebin, or your own web server.

    You can also import Dhall expressions from environment variables, too:

    >>> System.Environment.setEnv "FOO" "1"
    >>> input auto "env:FOO" :: IO Natural
    1
    

    You can import types, too. For example, we can change our ./bar file to:

    -- ./bar.dhall
    [3.0, 4.0, 5.0] : List ./type.dhall

    ... then specify the type in a separate file:

    -- ./type.dhall
    Double

    ... and everything still type checks:

    $ ./example
    Example {foo = 1, bar = [3.0,4.0,5.0]}

    Note that imports should be terminated by whitespace or parentheses otherwise you will get either an import error or a parse error, like this:

    >>> writeFile "baz" "2.0"
    >>> input auto "./baz: Double" :: IO Double
    *** Exception: 
    ↳ ./baz: 
    ...
    ...Error...: Missing file ...baz:
    ...
    

    This is because the parser thinks that ./baz: is a single token due to the missing whitespace before the colon and tries to import a file named ./baz:, which does not exist. To fix the problem we have to add a space after ./baz:

    >>> input auto "./baz : Double" :: IO Double
    2.0
    

    Exercise: There is a not function hosted online here:

    https://prelude.dhall-lang.org/Bool/not

    Visit that link and read the documentation. Then try to guess what this code returns:

    >>> input auto "https://prelude.dhall-lang.org/Bool/not https://raw.githubusercontent.com/dhall-lang/dhall-haskell/18e4e9a18dc53271146df3ccf5b4177c3552236b/examples/True" :: IO Bool
    ???

    Run the code to test your guess

    Lists

    You can store 0 or more values of the same type in a list, like this:

    [1, 2, 3]

    Every list can be followed by the type of the list. The type annotation is required for empty lists but optional for non-empty lists. You will get a type error if you provide an empty list without a type annotation:

    >>> input auto "[]" :: IO (Vector Natural)
    *** Exception:
    ...Error...: An empty list requires a type annotation
    ...
    1│ []
    ...
    (input):1:1
    

    Also, list elements must all have the same type. You will get an error if you try to store elements of different types in a list:

    >>> input auto "[1, True, 3]" :: IO (Vector Natural)
    *** Exception:
    ...Error...: List elements should all have the same type
    ...
    - Natural
    + Bool
    ...
    1│     True
    ...
    (input):1:5
    ...
    

    Exercise: Replace the "???" with an expression that successfully decodes to the specified type:

    >>> input auto "???" :: IO (Vector (Vector Natural))

    Optional values

    Optional values are either of the form Some value or None type.

    For example, these are valid Optional values:

    Some 1
    
    None Natural

    ... which both have type Optional Natural

    An Optional corresponds to Haskell's Maybe type for decoding purposes:

    >>> input auto "Some 1" :: IO (Maybe Natural)
    Just 1
    >>> input auto "None Natural" :: IO (Maybe Natural)
    Nothing
    

    Exercise: Replace the "???" with an expression that successfully decodes to the specified type:

    >>> input auto "???" :: IO (Maybe (Maybe (Maybe Natural)))

    Records

    Record literals are delimited by curly braces and their fields are separated by commas. For example, this is a valid record literal:

    { foo = True
    , bar = 2
    , baz = 4.2
    }

    A record type is like a record literal except instead of specifying each field's value we specify each field's type. For example, the preceding record literal has the following record type:

    { foo : Bool
    , bar : Natural
    , baz : Double
    }

    If you want to specify an empty record literal, you must use {=}, which is special syntax reserved for empty records. If you want to specify the empty record type, then you use {}. If you forget which is which you can always ask the dhall compiler to remind you of the type for each one:

    $ dhall type <<< '{=}'
    {}
    $ dhall type <<< '{}'
    Type

    This is our first use of the dhall command-line tool (provided by this package), which provides a type subcommand for inferring an expression's type. By default the tool reads the expression on standard input and outputs the type to standard output.

    Note that <<< is a feature specific to the Bash shell to feed a string to a command's standard input. If you are using another shell, then you can instead do this:

    $ echo '{=}' | dhall type
    {}

    Exercise: Use the dhall type command to infer the type of this record:

    -- ./nested.dhall
    { foo = 1
    , bar =
        { baz = 2.0
        , qux = True
        }
    }

    You can specify nested fields using dot-separated keys, like this:

    { foo = 1, bar.baz = 2.0, bar.qux = True }

    ... which is equivalent to:

    { foo = 1, bar = { baz = 2.0, qux = True } }

    You can also access a field of a record using the following syntax:

    record.fieldName

    ... which means to access the value of the field named fieldName from the record. For example:

    >>> input auto "{ foo = True, bar = 2, baz = 4.2 }.baz" :: IO Double
    4.2
    

    ... and you can project out multiple fields into a new record using this syntax:

    someRecord.{ field₀, field₁, … }

    For example:

    $ dhall <<< '{ x = 1, y = True, z = "ABC" }.{ x, y }'
    { x = 1, y = True }

    This is our first example of using the dhall command-line tool with no subcommand (like type), which evaluates the provided expression. By default, this reads the expression on standard input and outputs the evaluated result on standard output.

    Exercise: Save the above ./nested.dhall file and then try to access the value of the baz field. Test that this works by interpreting your code using the dhall command.

    You can convert a record to a list of key-value pairs (a.k.a. a "Map") by using the toMap keyword. For example:

    $ dhall <<< 'toMap { foo = 1, bar = 2 }'
    [ { mapKey = "bar", mapValue = 2 }, { mapKey = "foo", mapValue = 1 } ]

    This conversion only works if all field of the record have the same type. This comes in handy when you need to convert a Dhall record to the Dhall equivalent of a homogeneous map (i.e. Haskell's Data.Map).

    Functions

    The Dhall programming language also supports user-defined anonymous functions. For example, we can save the following anonymous function to a file:

    -- ./makeBools.dhall
    \(n : Bool) ->
        [ n && True, n && False, n || True, n || False ]

    ... or we can use Dhall's support for Unicode characters to use λ (U+03BB) instead of \ and (U+2192) instead of -> (for people who are into that sort of thing):

    $ -- ./makeBools.dhall
    λ(n : Bool) →
        [ n && True, n && False, n || True, n || False ]
    <Ctrl-D>

    You can read this as a function of one argument named n that has type Bool. This function returns a List of Bools. Each element of the List depends on the input argument named n.

    The (ASCII) syntax for anonymous functions resembles the syntax for anonymous functions in Haskell. The only difference is that Dhall requires you to annotate the type of the function's input.

    You can import this function into Haskell, too:

    >>> makeBools <- input auto "./makeBools.dhall" :: IO (Bool -> Vector Bool)
    >>> makeBools True
    [True,False,True,True]
    

    The reason this works is that there is an FromDhall instance for simple functions:

    instance (ToDhall a, FromDhall b) => FromDhall (a -> b)

    Thanks to currying, this instance works for functions of multiple simple arguments:

    >>> dhallAnd <- input auto "λ(x : Bool) → λ(y : Bool) → x && y" :: IO (Bool -> Bool -> Bool)
    >>> dhallAnd True False
    False
    

    However, you can't convert anything more complex than that (like a polymorphic or higher-order function). You will need to apply those functions to their arguments within Dhall before converting their result to a Haskell value.

    Just like FromDhall, you can derive ToDhall for user-defined data types:

    {-# LANGUAGE DeriveAnyClass    #-}
    {-# LANGUAGE DeriveGeneric     #-}
    {-# LANGUAGE OverloadedStrings #-}
    
    module Main where
    
    import Dhall
    
    data Example0 = Example0 { foo :: Bool, bar :: Bool }
        deriving (Generic, ToDhall)
    
    main = do
        f <- input auto "λ(r : { foo : Bool, bar : Bool }) → r.foo && r.bar"
        print (f (Example0 { foo = True, bar = False }) :: Bool)
        print (f (Example0 { foo = True, bar = True  }) :: Bool)

    The above program prints:

    False
    True

    Compiler

    We can also test our makeBools function directly from the command line. This library comes with a command-line executable program named dhall that you can use to both type-check files and convert them to a normal form. Our compiler takes a program on standard input and then prints the program's type to standard error followed by the program's normal form to standard output:

    $ dhall --annotate <<< './makeBools.dhall'
    (λ(n : Bool) → [ n, False, True, n ]) : ∀(n : Bool) → List Bool

    The --annotate flag adds a type signature to the output to let us know what type the interpreter inferred for our expression. The type signature is ∀(n : Bool) → List Bool which says that makeBools is a function of one argument named n that has type Bool and the function returns a List of Bools. The (U+2200) symbol is shorthand for the ASCII forall keyword:

    ∀(x : a) → b        -- This type ...
    
    forall (x : a) → b  -- ... is the same as this type

    ... and Dhall's forall keyword behaves the same way as Haskell's forall keyword for input values that are Types:

    forall (x : Type) → b  -- This Dhall type ...
    forall x . b           -- ... is the same as this Haskell type

    The part where Dhall differs from Haskell is that you can also use /forall to give names to non-Type arguments (such as the first argument to makeBools).

    This expression is our program's normal form:

    λ(n : Bool) → [ n, False, True, n ]

    ... and the interpreter was able to simplify our expression by noting that:

    • n && True  = n
    • n && False = False
    • n || True  = True
    • n || False = n

    To apply a function to an argument you separate the function and argument by whitespace (just like Haskell):

    f x

    You can read the above as "apply the function f to the argument x". This means that we can "apply" our ./makeBools function to a Bool argument like this:

    $ dhall <<< './makeBools.dhall True'
    [True, False, True, True]

    Remember that file paths are synonymous with their contents, so the above code is exactly equivalent to:

    $ dhall <<< '(λ(n : Bool) → [n && True, n && False, n || True, n || False]) True'
    [True, False, True, True]

    When you apply an anonymous function to an argument, you substitute the "bound variable" with the function's argument:

       Bound variable
       ⇩
    (λ(n : Bool) → ...) True
                        ⇧
                        Function argument

    So in our above example, we would replace all occurrences of n with True, like this:

    -- If we replace all of these `n`s with `True` ...
    [n && True, n && False, n || True, n || False]
    
    -- ... then we get this:
    [True && True, True && False, True || True, True || False]
    
    -- ... which reduces to the following normal form:
    [True, False, True, True]

    Now that we've verified that our function type checks and works, we can use the same function within Haskell:

    >>> input auto "./makeBools.dhall True" :: IO (Vector Bool)
    [True,False,True,True]
    

    Exercise: Create a file named getFoo that is a function of the following type:

    ∀(r : { foo : Bool, bar : Text }) → Bool

    This function should take a single input argument named r that is a record with two fields. The function should return the value of the foo field.

    Exercise: Use the dhall command to infer the type of the function you just created and verify that your function has the correct type

    Exercise: Use the dhall command to apply your function to a sample record

    Strings

    Dhall supports ordinary string literals with Haskell-style escaping rules:

    $ dhall
    "Hello, \"world\"!"
    <Ctrl-D>
    "Hello, \"world\"!"

    ... and Dhall also supports Nix-style multi-line string literals:

    $ dhall
    ''
        #!/bin/bash
        
        echo "Hi!"
    ''
    <Ctrl-D>
    "\n#!/bin/bash\n\necho \"Hi!\"\n"

    These "double single quote strings" ignore all special characters, with one exception: if you want to include a '' in the string, you will need to escape it with a preceding ' (i.e. use ''' to insert '' into the final string).

    These strings also strip leading whitespace using the same rules as Nix. Specifically: "it strips from each line a number of spaces equal to the minimal indentation of the string as a whole (disregarding the indentation of empty lines)."

    You can also interpolate expressions into strings using ${...} syntax. For example:

    $ dhall
    let name = "John Doe"
    let age  = 21
    in  "My name is ${name} and my age is ${Natural/show age}"
    <Ctrl-D>
    "My name is John Doe and my age is 21"

    Note that you can only interpolate expressions of type Text

    If you need to insert a "${" into a string without interpolation then use "''${" (same as Nix)

    ''
        for file in *; do
          echo "Found ''${file}"
        done
    ''

    Combine

    You can combine two records, using either the (//) operator or the (/\) operator.

    The (//) operator (or (⫽) U+2AFD) combines the fields of both records, preferring fields from the right record if they share fields in common:

    $ dhall
    { foo = 1, bar = "ABC" } // { baz = True }
    <Ctrl-D>
    { bar = "ABC", baz = True, foo = 1 }
    $ dhall
    { foo = 1, bar = "ABC" } ⫽ { bar = True }  -- Fancy unicode
    <Ctrl-D>
    { bar = True, foo = 1 }

    Note that the order of record fields does not matter. The compiler automatically sorts the fields.

    If you need to set or add a deeply nested field you can use the with keyword, like this:

    $ dhall <<< '{ x.y = 1 } with x.z = True'
    { x = { y = 1, z = True } }
    $ dhall <<< '{ x.y = 1 } with x.y = 2'
    { x.y = 2 }

    The with keyword is syntactic sugar for the // operator which follows these rules:

    -- Nested case
    record with k.ks… = value  ⇒  record // { k = record.k with ks… = value }
    
    -- Base case
    record with k = value      ⇒  record // { k = value }

    The (/\) operator (or (∧) U+2227) also lets you combine records, but behaves differently if the records share fields in common. The operator combines shared fields recursively if they are both records:

    $ dhall
    { foo = { bar = True }, baz = "ABC" } /\ { foo = { qux = 1.0 } }
    <Ctrl-D>
    { baz = "ABC", foo = { bar = True, qux = 1.0 } }

    ... but fails with a type error if either shared field is not a record:

    $ dhall
    { foo = 1, bar = "ABC" } ∧ { foo = True }
    <Ctrl-D>
    Use "dhall --explain" for detailed errors
    
    Error: Field collision
    
    { foo = 1, bar = "ABC" } ∧ { foo = True }
    
    (input):1:1

    Exercise: Combine any record with the empty record. What do you expect to happen?

    You can analogously combine record types using the //\\ operator (or (⩓) U+2A53):

    $ dhall
    { foo : Natural } ⩓ { bar : Text }
    <Ctrl-D>
    { foo : Natural, bar : Text }

    ... which behaves the exact same, except at the type level, meaning that the operator descends recursively into record types:

    $ dhall
    { foo : { bar : Text } } ⩓ { foo : { baz : Bool }, qux : Natural }
    <Ctrl-D>
    { foo : { bar : Text, baz : Bool }, qux : Natural }

    Let expressions

    Dhall also supports let expressions, which you can use to define intermediate values in the course of a computation, like this:

    $ dhall
    let x = "ha" in x ++ x
    <Ctrl-D>
    "haha"

    You can also annotate the types of values defined within a let expression, like this:

    $ dhall
    let x : Text = "ha" in x ++ x
    <Ctrl-D>
    "haha"

    You need to nest let expressions if you want to define more than one value in this way:

    $ dhall
    let x = "Hello, "
    let y = "world!"
    in  x ++ y
    <Ctrl-D>
    "Hello, world!"

    Dhall is whitespace-insensitive, so feel free to format things over multiple lines or indent in any way that you please.

    If you want to define a named function, just give a name to an anonymous function:

    $ dhall
    let twice = λ(x : Text) → x ++ x in twice "ha"
    <Ctrl-D>
    "haha"

    Unlike Haskell, Dhall does not support function arguments on the left-hand side of the equals sign, so this will not work:

    $ dhall
    let twice (x : Text) = x ++ x in twice "ha"
    <Ctrl-D>
    Error: Invalid input
    
    (input):1:11:
      |
    1 | let twice (x : Text) = x ++ x in twice "ha"
      |           ^
    unexpected '('
    expecting ':', '=', or the rest of label

    The error message says that Dhall expected either a (:) (i.e. the beginning of a type annotation) or a (=) (the beginning of the assignment) and not a function argument.

    You can also use let expressions to rename imports, like this:

    $ dhall
    let not = https://prelude.dhall-lang.org/Bool/not
    in  not True
    <Ctrl-D>
    False

    ... or to define synonyms for types:

    $ dhall <<< 'let Name : Type = Text in [ "John", "Mary" ] : List Name'
    List Text
    
    [ "John", "Mary" ]

    Exercise: What do you think the following code will normalize to?

    let x = 1
    let x = 2
    in  x

    Test your guess using the dhall compiler

    Exercise: Now try to guess what this code will normalize to:

    let x = "ha"
    let x = x ++ "ha"
    in  x

    Exercise: What about this code?

    let x = x ++ "ha"
    in  x

    Defaults

    For records with a large number of defaultable fields you can use the :: operator to auto-complete a record. For example:

    let greet =
              \(args : { greeting : Text, name : Text })
          ->  "${args.greeting}, ${args.name}!"
    
    let Greeting =
          { Type = { greeting : Text, name : Text }
          , default = { greeting = "Hello", name = "John" }
          }
    
    in  ''
        ${greet Greeting::{=}}
        ${greet Greeting::{ greeting = "Hola" }}
        ${greet Greeting::{ name = "Jane" }}
        ${greet Greeting::{ greeting = "Hola", name = "Jane" }}
        ''

    This operator is syntactic sugar. Specifically an expression of the form:

    T::r

    ... is equivalent to:

    (T.default // r) : T.Type

    So, for example:

    Greeting::{ greeting = "Hola" }

    ... is the same thing as:

    (Greeting.default // { greeting = "Hola" }) : Greeting.Type

    ... which is the same thing as:

    ({ greeting = "Hello", name = "John" } // { greeting = "Hola" }) : { greeting : Text, name : Text }

    ... which is the same thing as:

    { greeting = "Hola", name = "John" }

    Unions

    A union is a value that can be one of many alternative types of values. For example, the following union type:

    < Left : Natural | Right : Bool >

    ... represents a value that can be either a Natural or a Bool value. If you are familiar with Haskell these are exactly analogous to Haskell's "sum types" and the above type is equivalent to Either Natural Bool.

    Each alternative is associated with a tag that distinguishes that alternative from other alternatives. In the above example, the Left tag is used for the Natural alternative and the Right tag is used for the Bool alternative.

    You can specify the value of a union constructor like this:

    let Union = < Left : Natural | Right : Bool>
    
    in  [ Union.Left 0, Union.Right True ]

    In other words, you can access a union constructor as a field of a union type and use that constructor to wrap a value of a type appropriate for that alternative. In the above example, the Left constructor can wrap a Natural value and the Right constructor can wrap a Bool value. We can also confirm that by inspecting their type:

    $ echo '< Left : Natural | Right : Bool>' > ./Union
    $ dhall --annotate <<< '(./Union).Left'
      < Left : Natural | Right : Bool >.Left
    : ∀(Left : Natural) → < Left : Natural | Right : Bool >
    $ dhall --annotate <<< '(./Union).Right'
      < Left : Natural | Right : Bool >.Right
    : ∀(Right : Bool) → < Left : Natural | Right : Bool >

    In other words, the Left constructor is a function from a Natural to a value of our Union type and the Right constructor is a separate function from a Bool to that same Union type.

    You can consume a union using the built-in merge function. For example, suppose we want to convert our union to a Bool but we want to behave differently depending on whether or not the union is a Natural wrapped in the Left constructor or a Bool wrapped in the Right constructor . We would write:

    $ cat > process <<EOF
        λ(union : < Left : Natural | Right : Bool >)
    →   let handlers =
                { Left  = Natural/even  -- Natural/even is a built-in function
                , Right = λ(b : Bool) → b
                }
    in  merge handlers union
    EOF

    Now our ./process function can handle both alternatives:

    $ dhall <<< './process ((./Union).Left 3)'
    False
    $ dhall <<< './process ((./Union).Right True)'
    True

    Every merge has the following form:

    merge handlers union [ : type ]

    ... where:

    • union is the union you want to consume
    • handlers is a record with one function per alternative of the union
    • type is an optional declared result type of the merge

    The merge function selects which function to apply from the record based on which alternative the union selects:

    merge { Foo = f, ... } (< … >.Foo x) = f x

    So, for example:

    merge { Left = Natural/even, Right = λ(b : Bool) → b } (< Left : Natural | Right : Bool >.Left 3)
        = Natural/even 3
        = False

    ... and similarly:

    merge { Left = Natural/even, Right = λ(b : Bool) → b } (< Left : Natural | Right : Bool >.Right True)
        = (λ(b : Bool) → b) True
        = True

    Notice that each handler has to return the same type of result (Bool in this case).

    You can also store more than one value within alternatives using Dhall's support for anonymous records. You can nest an anonymous record within a union such as in this type:

    < Empty : {} | Person : { name : Text, age : Natural } >

    You can even go a step further and omit the type of an alternative if it stores no data, like this:

    < Empty | Person : { name : Text, age : Natural } >

    The above Dhall type resembles the following equivalent Haskell data type:

    data Example = Empty | Person { name :: Text, age :: Text }

    Empty alternatives like Empty require no argument:

    let Example = < Empty | Person : { name : Text, age : Natural } >
    
    in  [ Example.Empty  -- Note the absence of any argument to `Empty`
        , Example.Person { name = "John", age = 23 }
        , Example.Person { name = "Amy" , age = 25 }
        ]

    ... and when you merge an empty alternative the correspond handler takes no argument:

        λ(x : < Empty | Person : { name : Text, age : Natural } >)
    →   merge
        {   Empty = "Unknown"  -- Note the absence of a `λ`
    
        ,   Person =
                λ(person : { name : Text, age : Natural })
            →   "Name: ${person.name}, Age: ${Natural/show person.age}"
        }
        x

    Exercise: Create a list of the following type with at least one element per alternative:

    List < Left : Natural | Right : Double >

    Polymorphic functions

    The Dhall language supports defining polymorphic functions (a.k.a. "generic" functions) that work on more than one type of value. However, Dhall differs from Haskell by not inferring the types of these polymorphic functions. Instead, you must be explicit about what type of value the function is specialized to.

    Take, for example, Haskell's identity function named id:

    id :: a -> a
    id = \x -> x

    The identity function is polymorphic, meaning that id works on values of different types:

    >>> id 4
    4
    >>> id True
    True
    

    The equivalent function in Dhall is:

    λ(a : Type) → λ(x : a) → x

    Notice how this function takes two arguments instead of one. The first argument is the type of the second argument.

    Let's illustrate how this works by actually using the above function:

    $ echo "λ(a : Type) → λ(x : a) → x" > id

    Let's ask the interpreter for the type of this function: the first line:

    $ dhall type <<< './id'
    ∀(a : Type) → ∀(x : a) → a

    You can read the type (∀(a : Type) → ∀(x : a) → a) as saying: "This is the type of a function whose first argument is named a and is a Type. The second argument is named x and has type a (i.e. the value of the first argument). The result also has type a."

    This means that the type of the second argument changes depending on what type we provide for the first argument. When we apply ./id to Natural, we create a function that expects an Natural argument:

    $ dhall type <<< './id Natural'
    ∀(x : Natural) → Natural

    Similarly, when we apply ./id to Bool, we create a function that expects a Bool argument:

    $ dhall type <<< './id Bool'
    ∀(x : Bool) → Bool

    We can then supply the final argument to each of those functions to show that they both work on their respective types:

    $ dhall <<< './id Natural 4'
    4
    $ dhall <<< './id Bool True'
    True

    Built-in functions can also be polymorphic, too. For example, we can ask the compiler for the type of List/reverse, the function that reverses a list:

    $ dhall --annotate
    List/reverse
    <Ctrl-D>
    List/reverse : ∀(a : Type) → List a → List a

    The first argument to List/reverse is the type of the list to reverse:

    $ dhall
    List/reverse Bool
    <Ctrl-D>
    List/reverse Bool : List Bool → List Bool

    ... and the second argument is the list to reverse:

    $ dhall
    List/reverse Bool [True, False]
    <Ctrl-D>
    [False, True]

    Note that the second argument has no name. This type:

    ∀(a : Type) → List a → List a

    ... is equivalent to this type:

    ∀(a : Type) → ∀(_ : List a) → List a

    In other words, if you don't see the symbol surrounding a function argument type then that means that the name of the argument is "_". This is true even for user-defined functions:

    $ dhall type <<< 'λ(_ : Text) → 1'
    Text → Natural

    The type Text → Natural is the same as ∀(_ : Text) → Natural

    Exercise : Translate Haskell's flip function to Dhall

    Total

    Dhall is a total programming language, which means that Dhall is not Turing-complete and evaluation of every Dhall program is guaranteed to eventually halt. There is no upper bound on how long the program might take to evaluate, but the program is guaranteed to terminate in a finite amount of time and not hang forever.

    This guarantees that all Dhall programs can be safely reduced to a normal form where as many functions have been evaluated as possible. In fact, Dhall expressions can be evaluated even if all function arguments haven't been fully applied. For example, the following program is an anonymous function:

    $ dhall
    \(n : Bool) -> 10 * 10
    <Ctrl-D>
    λ(n : Bool) → 100

    ... and even though the function is still missing the first argument named n the compiler is smart enough to evaluate the body of the anonymous function ahead of time before the function has even been invoked.

    We can use the map function from the Prelude to illustrate an even more complex example:

    $ dhall
        let List/map = https://prelude.dhall-lang.org/List/map
    in  λ(f : Natural → Natural) → List/map Natural Natural f [1, 2, 3]
    <Ctrl-D>
    λ(f : Natural → Natural) → [f 1, f 2, f 3]

    Dhall can apply our function to each element of the list even before we specify which function to apply.

    The language will also never crash or throw any exceptions. Every computation will succeed and produce something, even if the result might be an Optional value:

    $ dhall <<< 'List/head Natural ([] : List Natural)'
    None Natural

    Exercise: The Dhall Prelude provides a replicate function which you can find here:

    https://prelude.dhall-lang.org/List/replicate

    Test what the following Dhall expression normalizes to:

    let replicate = https://prelude.dhall-lang.org/List/replicate
    in  replicate 10

    Exercise: If you have a lot of spare time, try to "break the compiler" by finding an input expression that crashes or loops forever (and file a bug report if you succeed).

    Assertions

    You can add compile-time assertions which you can use to add tests to your code. For example, we can add some tests to our not function like this:

    let not
        : Bool → Bool
        = λ(b : Bool) → b == False
    
    let example0 = assert : not False === True
    
    let example1 = assert : not True === False
    
    in  not

    The expression assert : not False == True is a type-checking assertion that two expressions have the same normal form. If the two expressions differ then type-checking rejects the code.

    For example, suppose that we change the example to add an incorrect assertion:

    -- ./test.dhall
    
    let not
        : Bool → Bool
        = λ(b : Bool) → b == False
    
    let example0 = assert : not False === True
    
    let example1 = assert : not True === True  -- Oops!
    
    in  not

    The type-checker then rejects the assertion with the following error message:

    $ dhall <<< './test.dhall' 
    
    ↳ ./test.dhall
    
    Error: Assertion failed
    
    - False
    + True
    
    7│                assert : not True === True -- Oops!
    8│ 
    
    ./test.dhall:7:16
    
    1│ ./test.dhall
    
    (input):1:1

    You can compare expressions that contain variables, too, which is equivalent to symbolic reasoning:

    $ dhall <<< '\(n : Natural) -> assert : n === (n + 0)'
    λ(n : Natural) → assert : n ≡ n

    Dhall accepts this because the language has built-in support for normalizing n + 0 to n, so both sides of the comparison normalize to the same value: n.

    Note that this sort of symbolic reasoning is limited and can only detect equality of normal forms. Some equivalent expressions will be rejected if they don't share the same normal form, such as these:

    $ dhall <<< '\(n : Natural) -> assert : Natural/even (n + n) === True'
    
    Use "dhall --explain" for detailed errors
    
    n : Natural
    
    Error: Assertion failed
    
    - … …
    + True
    
    1│                   assert : Natural/even (n + n) === True
    
    (input):1:19

    Here the interpreter is not smart enough to simplify Natural/even (n + n) to True so the assertion fails.

    If you prefer to use Unicode, then the Unicode equivalent of === is (U+2261).

    Headers

    Sometimes you would like to provide additional request headers when importing Dhall expressions from URLs. For example, you might want to provide an authorization header or specify the expected content type.

    Dhall URL imports let you add or modify request headers with the using keyword:

    https://example.com/example.dhall using ./headers

    ... where you can replace ./headers with any import that points to a Dhall expression of the following type:

    List { header : Text, value : Text }

    For example, if you needed to specify the content type correctly in order to download the file, then your ./headers file might look like this:

    $ cat headers
    [ { header = "Accept", value = "application/dhall" } ]

    ... or if you needed to provide an authorization token to access a private GitHub repository, then your headers could look like this:

    [ { header = "Authorization", value = "token ${env:GITHUB_TOKEN as Text}" } ]

    This would read your GitHub API token from the GITHUB_TOKEN environment variable and supply that token in the authorization header.

    You cannot inline the headers within the same file as the URL. You must provide them as a separate import. That means that this is not legal code:

    http://example.com/example.dhall using [ { header = "Accept", value = "application/dhall" } ]  -- NOT legal

    Dhall will forward imports if you import an expression from a URL that contains a relative import. For example, if you import an expression like this:

    http://example.com/example.dhall using ./headers

    ... and http://example.com/example.dhall contains a relative import of ./foo then Dhall will import http://example.com/foo using the same ./headers file.

    Import integrity

    Sometimes you want to use share code while still ensuring that the imported value never changes and can't be corrupted by a malicious attacker. Dhall provides built-in support for hashing imported values to verify that their value never changes

    For example, suppose you save the following two files:

    $ cat ./foo
    ./bar sha256:d60d8415e36e86dae7f42933d3b0c4fe3ca238f057fba206c7e9fbf5d784fe15
    $ cat ./bar
    ./baz
    $ cat ./baz
    1

    The first file named ./foo contains an example of an integrity check. You can add sha256:XXX after any import (such as after ./bar), where XXX is an expected 64-character sha256 hash of the Dhall value. To be precise, the hash represents a sha256 hash of the UTF-8 encoding of a canonical text representation of the fully resolved and normalized abstract syntax tree of the imported expression.

    Dhall will verify that the expected hash matches the actual hash of the imported Dhall value and reject the import if there is a hash mismatch:

    $ dhall <<< './foo'
    Natural
    
    1

    Any import protected by a semantic integrity check is automatically cached locally underneath either ~/.cache/dhall/1220${HASH} or ${XDG_CACHE_HOME}/dhall/1220${HASH} if you define the XDG_CACHE_HOME environment variable.

    For example, after you import ./foo the contents of `./bar` are locally cached in a fully-evaluated and binary-encoded form which you can inspect by running:

    $ dhall decode < ~/.cache/dhall/1220d60d8415e36e86dae7f42933d3b0c4fe3ca238f057fba206c7e9fbf5d784fe15
    1

    Subsequent attempts to resolve the same import will automatically retrieve the import from the cache. This matters less for local imports, but comes in handy for remote imports to avoid redownloading them.

    The local cache takes precedence when resolving imports, so changing the original import afterwards will have no affect until you update the hash. From Dhall's point of view, the hash is the true address and the file path is just a suggestion for how to obtain the import if it's not already cached.

    You can disable the cache by setting XDG_CACHE_HOME to `/var/empty` (an empty and unwritable directory), like this:

    $ XDG_CACHE_HOME=/var/empty dhall <<< './foo'
    Natural
    
    1

    We'll use this trick to test changes to the protected import in the following examples.

    Now, suppose you add a comment to the ./bar file:

    $ cat ./bar
    -- This comment does not change the hash
    ./baz

    ... then ./foo will still successfully import ./bar because the hash only depends on the normalized value and does not depend on meaningless changes to whitespace or comments:

    $ XDG_CACHE_HOME=/var/empty dhall <<< './foo'  # This still succeeds
    Natural
    
    1

    You can compute the Hash for any import by using the hash subcommand of this package. For example:

    $ dhall hash <<< './bar'
    sha256:d60d8415e36e86dae7f42933d3b0c4fe3ca238f057fba206c7e9fbf5d784fe15

    Now suppose that you actually change the value of the ./baz file:

    $ cat ./baz
    2

    ... then the ./foo file will fail to import ./bar, even though the text of the ./bar file technically never changed:

    XDG_CACHE_HOME=/var/empty dhall <<< './foo'
    
    Error: Import integrity check failed
    
    Expected hash:
    
    ↳ d60d8415e36e86dae7f42933d3b0c4fe3ca238f057fba206c7e9fbf5d784fe15
    
    Actual hash:
    
    ↳ 4caf97e8c445d4d4b5c5b992973e098ed4ae88a355915f5a59db640a589bc9cb

    This is because the ./bar file now represents a new value (2 instead of 1), even though the text of the ./bar file is still the same. Since the value changed the hash must change as well. However, we could change ./baz to:

    $ cat baz
    if True then 1 else 2

    ... and the import would now succeed again because the final result is 1.

    The integrity hash ensures that your import's final meaning can never change, so an attacker can never compromise an imported value protected by a hash unless they can break SHA-256 encryption. The hash not only protects the file that you immediately import, but also protects every transitive import as well.

    You can also safely refactor your imported dependencies knowing that the refactor will not change your hash so long as your refactor is behavior-preserving. This provides an easy way to detect refactoring errors that you might accidentally introduce. The hash not only protects you from attackers, but also protects against human error, too!

    If you have a file which either doesn't already use hashed imports, or you changed some of the imports and want to update the hashes you can use the freeze command to either add or update hashes:

    $ cat foo.dhall
    let replicate = https://prelude.dhall-lang.org/List/replicate
    in  replicate 5
    $
    $ dhall freeze --inplace ./foo.dhall
    $ cat ./foo.dhall
    let replicate =
          https://prelude.dhall-lang.org/List/replicate sha256:d4250b45278f2d692302489ac3e78280acb238d27541c837ce46911ff3baa347
    
    in  replicate 5

    Raw text

    Sometimes you want to import the contents of a raw text file as a Dhall value of type Text. For example, one of the fields of a record might be the contents of a software license:

    { package = "dhall"
    , author  = "Gabriel Gonzalez"
    , license = ./LICENSE
    }

    Normally if you wanted to import a text file you would need to wrap the contents of the file in double single-quotes, like this:

    $ cat LICENSE
    ''
    Copyright (c) 2017 Gabriel Gonzalez
    All rights reserved.
    
    ...
    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
    ''

    ... which does not work well if you need to reuse the same text file for other programs

    However, Dhall supports importing a raw text file by adding as Text to the end of the import, like this:

    { package = "dhall"
    , author  = "Gabriel Gonzalez"
    , license = ./LICENSE as Text
    }

    ... and then you can use the original text file unmodified:

    $ cat LICENSE
    Copyright (c) 2017 Gabriel Gonzalez
    All rights reserved.
    
    ...
    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

    Formatting code

    A format subcommand is also available that you can use to automatically format Dhall expressions. For example, we can take the following unformatted Dhall expression:

    $ cat ./unformatted
    λ(a : Type) → λ(kvss : List (List { index : Natural, value : a })) → 
    List/build { index : Natural, value : a } (λ(list : Type) → λ(cons : { 
    index : Natural, value : a } → list → list) → λ(nil : list) → 
    (List/fold (List { index : Natural, value : a }) kvss { count : Natural, diff : 
    Natural → list } (λ(kvs : List { index : Natural, value : a }) → λ(y : { 
    count : Natural, diff : Natural → list }) → { count = y.count + List/length 
    { index : Natural, value : a } kvs, diff = λ(n : Natural) → List/fold { 
    index : Natural, value : a } kvs list (λ(kvOld : { index : Natural, value : a 
    }) → λ(z : list) → cons { index = kvOld.index + n, value = kvOld.value } 
    z) (y.diff (n + List/length { index : Natural, value : a } kvs)) }) { count = 
    0, diff = λ(_ : Natural) → nil }).diff 0)

    ... and run the expression through the formatter:

    $ dhall format < ./unformatted
      λ(a : Type)
    → λ(kvss : List (List { index : Natural, value : a }))
    → List/build
        { index : Natural, value : a }
        (   λ(list : Type)
          → λ(cons : { index : Natural, value : a } → list → list)
          → λ(nil : list)
          → ( List/fold
                (List { index : Natural, value : a })
                kvss
                { count : Natural, diff : Natural → list }
                (   λ(kvs : List { index : Natural, value : a })
                  → λ(y : { count : Natural, diff : Natural → list })
                  → { count =
                        y.count + List/length { index : Natural, value : a } kvs
                    , diff =
                          λ(n : Natural)
                        → List/fold
                            { index : Natural, value : a }
                            kvs
                            list
                            (   λ(kvOld : { index : Natural, value : a })
                              → λ(z : list)
                              → cons
                                  { index = kvOld.index + n, value = kvOld.value }
                                  z
                            )
                            ( y.diff
                                (n + List/length { index : Natural, value : a } kvs)
                            )
                    }
                )
                { count = 0, diff = λ(_ : Natural) → nil }
            ).diff
              0
        )

    The executable formats expressions without resolving, type-checking, or normalizing them:

    $ dhall format
    let replicate = https://prelude.dhall-lang.org/List/replicate 
    in replicate 5 (List (List Natural)) (replicate 5 (List Natural) (replicate 5 Natural 1))
    <Ctrl-D>
    let replicate = https://prelude.dhall-lang.org/List/replicate
    
    in  replicate
          5
          (List (List Natural))
          (replicate 5 (List Natural) (replicate 5 Natural 1))

    You can also use the formatter to modify files in place using the --inplace flag (i.e. for formatting source code):

    $ dhall format --inplace ./unformatted

    Carefully note that the code formatter does not preserve all comments. Currently, the formatter only preserves two types of comments:

    • Leading comments at the beginning of the file
    • Comments within a let binding

    For example:

    $ dhall format
    {- This comment will be preserved by the formatter -}
    -- ... and this comment will be preserved, too
    {- This comment will *NOT* be preserved -} 1
    -- ... and this comment will also *NOT* be preserved
    <Ctrl-D>
    {- This comment will be preserved by the formatter -}
    -- ... and this comment will be preserved, too
    1

    Also:

    let {- This comment will be preserved -}
        x {- This comment will also be preserved-} =
        {- ... and this one will be preserved, too -}
          1
    
    in  x

    Note that you do not need to format the output of the dhall interpreter. The interpreter already automatically formats multi-line expressions, too:

    $ dhall
    let replicate = https://prelude.dhall-lang.org/List/replicate 
    in replicate 5 (List (List Natural)) (replicate 5 (List Natural) (replicate 5 Natural 1))
    <Ctrl-D>
    [ [ [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      ]
    , [ [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      ]
    , [ [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      ]
    , [ [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      ]
    , [ [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      , [ 1, 1, 1, 1, 1 ]
      ]
    ]

    Built-in functions

    Dhall is a restricted programming language that only supports simple built-in functions and operators. If you want to do anything fancier you will need to load your data into Haskell for further processing

    This section covers types, functions, and operators that are built into the language, meaning that you do not need to import any code to use them. Additionally, Dhall also comes with a Prelude (covered in the next section) hosted online that contains functions derived from these base utilities. The Prelude also re-exports all built-in functions for people who prefer consistency.

    The language provides built-in support for the following primitive types:

    • Bool values
    • Natural values
    • Integer values
    • Double values
    • Text values

    ... as well as support for the following derived types:

    • Lists of values
    • Optional values
    • Anonymous records
    • Anonymous unions

    You can find an up-to-date list of available built-in functions and operators here:

    Built-in types, functions, and operators

    Caveats

    Dhall differs in a few important ways from other programming languages, so you should keep the following caveats in mind:

    First, Dhall only supports addition and multiplication on Natural numbers (i.e. non-negative integers), which are not the same type of number as Integers (which can be negative). An Integer number is a number prefixed with either a + or - symbol whereas a Natural number has no leading sign. If you try to add or multiply two Integers you will get a type error:

    $ dhall
    +2 + +2
    <Ctrl-D>
    Use "dhall --explain" for detailed errors
    
    Error: ❰+❱ only works on ❰Natural❱s
    
    +2 + +2
    
    (input):1:1

    In fact, there are no built-in functions for Integers (or Doubles) other than Integer/show and Double/show. As far as the language is concerned they are opaque values that can only be shuffled around but not used in any meaningful way until they have been loaded into Haskell.

    Second, the equality (==) and inequality (!=) operators only work on Bools. You cannot test any other types of values for equality.

    However, you can extend the language with your own built-ins using the Haskell API, as described in the next section.

    Extending the language

    You can use the Haskell API to extend the Dhall configuration language with new built-in functions. This section contains a simple Haskell recipe to add a new Natural/equal built-in function of type:

    Natural/equal : Natural → Natural → Bool

    To do so, we:

    • extend the type-checking context to include the type of Natural/equal
    • extend the normalizer to evaluate all occurrences of Natural/equal

    ... like this:

    -- example.hs
    
    {-# LANGUAGE OverloadedStrings #-}
    
    module Main where
    
    import Dhall.Core (Expr(..), ReifiedNormalizer(..))
    
    import qualified Data.Text.IO
    import qualified Dhall
    import qualified Dhall.Context
    import qualified Lens.Family   as Lens
    
    main :: IO ()
    main = do
        text <- Data.Text.IO.getContents
    
        let startingContext = transform Dhall.Context.empty
              where
                transform = Dhall.Context.insert "Natural/equal" naturalEqualType
    
                naturalEqualType =
                    Pi "_" Natural (Pi "_" Natural Bool)
    
        let normalizer (App (App (Var "Natural/equal") (NaturalLit x)) (NaturalLit y)) =
                Just (BoolLit (x == y))
            normalizer _ =
                Nothing
    
        let inputSettings = transform Dhall.defaultInputSettings
              where
                transform =
                      Lens.set Dhall.normalizer      (Just (ReifiedNormalizer (pure . normalizer)))
                    . Lens.set Dhall.startingContext startingContext
    
        x <- Dhall.inputWithSettings inputSettings Dhall.auto text
    
        Data.Text.IO.putStrLn x

    Here is an example use of the above program:

    $ ./example <<< 'if Natural/equal 2 (1 + 1) then "Equal" else "Not equal"'
    Equal

    Note that existing Dhall tools that type-check expressions will reject expressions containing unexpected free variable such as Natural/equal:

    $ dhall <<< 'Natural/equal 2 (1 + 1)'
    
    Use "dhall --explain" for detailed errors
    
    Error: Unbound variable
    
    Natural/equal 
    
    (input):1:1

    You will need to either:

    The general guidelines for adding new built-ins to the language are:

    • Keep built-ins easy to implement across language bindings
    • Prefer general purpose built-ins or built-ins appropriate for the task of program configuration
    • Design built-ins to catch errors as early as possible (i.e. when type-checking the configuration)

    Substitutions

    Substitutions are another way to extend the language. Suppose we have the following Haskell datatype:

    data Result = Failure Integer | Success String
        deriving (Eq, Generic, Show)
    
    instance FromDhall Result

    We can use it in Dhall like that:

    -- example.dhall
    
    let Result = < Failure : Integer | Success Text >
    in Result.Failure 1

    Right now it is quite easy to keep these two definitions (the one in Haskell source and the one in the Dhall file) synchronized: If we implement a new feature in the Haskell source we update the corresponding type in the Dhall file. But what happens if our application is growing and our Result type contains e.g. 10 unions with possible types embedded in it? Maintaining the code will get tedious. Luckily we can extract the correct Dhall type from the Haskell definition:

    resultDecoder :: Dhall.Decoder Result
    resultDecoder = Dhall.auto
    
    resultType :: Expr Src Void
    resultType = Dhall.expected resultDecoder
    
    resultTypeString :: String
    resultTypeString = show $ pretty resultType

    Now we just have to inject that type into the Dhall code and we are done. One common way to do that is to wrap the import of example.dhall in a let expression:

    Dhall.input (Dhall.auto :: Dhall.Decoder Result) ("let Result = " ++ Data.Text.pack resultTypeString ++ " in ./example.dhall")

    Now we can omit the definition of Result in our example.dhall file. While this will work perfectly Dhall provides a cleaner solution for our "injection problem":

    {-# LANGUAGE OverloadedStrings #-}
    
    myexample :: IO Result
    myexample = let
       evaluateSettings = Lens.over Dhall.substitutions (Dhall.Map.insert "Result" resultType) Dhall.defaultEvaluateSettings
       in Dhall.inputFileWithSettings evaluateSettings resultDecoder "example.dhall"

    Substitutions are a simple Map mapping variables to expressions. The application of these substitutions reflect the order of the insertions in the substitution map:

    {-# LANGUAGE OverloadedStrings #-}
    
    substitute (Dhall.Core.Var "Foo") (Dhall.Map.fromList [("Foo", Dhall.Core.Var "Bar"), ("Bar", Dhall.Core.BoolLit True)])

    results in Dhall.Core.Var "Baz" since we first substitute "Foo" with "Bar" and then the resulting "Bar" with the final True.

    Notable differences to the other extensions of the builtin language:

    • This approach works well with the inputFile/inputFileWithSettings functions while the let-wrapping will not.
    • In contrast to the custom built-ins described above substitutions are made BEFORE the type-checking.

    Prelude

    There is also a Prelude available at:

    https://prelude.dhall-lang.org

    If you visit the above link you can browse the Prelude, which has a few subdirectories. For example, the Bool subdirectory has a not file, which you can reference using this URL:

    https://prelude.dhall-lang.org/Bool/not

    The not function is just a UTF8-encoded text file hosted online with the following contents

    $ $ dhall <<< 'https://prelude.dhall-lang.org/Bool/not as Text'
    ''
    {-
    Flip the value of a `Bool`
    -}
    let not : Bool → Bool = λ(b : Bool) → b == False
    
    let example0 = assert : not True ≡ False
    
    let example1 = assert : not False ≡ True
    
    in  not
    ''

    The file could have been much shorter, like this:

    λ(b : Bool) → b == False

    ... but all the functions exported from the Prelude try to be as self-documenting as possible by including:

    • the name of the function
    • the type of the function
    • documentation
    • tests

    The performance penalty for adding these helpful features is negligible, especially if you protect the import with a semantic integrity check because then the import would be cached compactly as λ(_ : Bool) → _ == False.

    You can use this not function either directly:

    $ dhall
    https://prelude.dhall-lang.org/Bool/not True
    <Ctrl-D>
    False

    ... or assign the URL to a shorter name:

    $ dhall
    let Bool/not = https://prelude.dhall-lang.org/Bool/not
    in  Bool/not True
    <Ctrl-D>
    False

    Some functions in the Prelude just re-export built-in functions for consistency and documentation, such as Prelude/Natural/even, which re-exports the built-in Natural/even function:

    $ dhall <<< 'https://prelude.dhall-lang.org/Natural/even as Text'
    ''
    {-
    Returns `True` if a number if even and returns `False` otherwise
    -}
    let even : Natural → Bool = Natural/even
    
    let example0 = assert : even 3 ≡ False
    
    let example1 = assert : even 0 ≡ True
    
    in  even
    ''

    You can also clone the Prelude locally to your filesystem if you prefer using `git clone`, like this:

    $ git clone https://github.com/dhall-lang/dhall-lang.git
    $ tree dhall-lang/Prelude
    dhall-lang/Prelude
    ├── Bool
    │   ├── and
    │   ├── build
    │   ├── even
    │   ├── fold
    │   ├── not
    │   ├── odd
    │   ├── or
    │   ├── package.dhall
    │   └── show
    ├── Double
    │   ├── package.dhall
    │   └── show
    ├── Function
    │   ├── compose
    │   └── package.dhall
    ├── Integer
    │   ├── package.dhall
    │   ├── show
    │   └── toDouble
    ├── JSON
    │   ├── Nesting
    │   ├── Tagged
    │   ├── Type
    │   ├── array
    │   ├── bool
    │   ├── keyText
    │   ├── keyValue
    │   ├── null
    │   ├── number
    │   ├── object
    │   ├── package.dhall
    │   ├── render
    │   └── string
    ├── List
    │   ├── all
    │   ├── any
    │   ├── build
    │   ├── concat
    │   ├── concatMap
    │   ├── default
    │   ├── empty
    │   ├── filter
    │   ├── fold
    │   ├── generate
    │   ├── head
    │   ├── indexed
    │   ├── iterate
    │   ├── last
    │   ├── length
    │   ├── map
    │   ├── null
    │   ├── package.dhall
    │   ├── replicate
    │   ├── reverse
    │   ├── shifted
    │   └── unzip
    ├── Location
    │   ├── Type
    │   └── package.dhall
    ├── Map
    │   ├── Entry
    │   ├── Type
    │   ├── empty
    │   ├── keyText
    │   ├── keyValue
    │   ├── keys
    │   ├── map
    │   ├── package.dhall
    │   └── values
    ├── Monoid
    ├── Natural
    │   ├── build
    │   ├── enumerate
    │   ├── equal
    │   ├── even
    │   ├── fold
    │   ├── greaterThan
    │   ├── greaterThanEqual
    │   ├── isZero
    │   ├── lessThan
    │   ├── lessThanEqual
    │   ├── odd
    │   ├── package.dhall
    │   ├── product
    │   ├── show
    │   ├── subtract
    │   ├── sum
    │   ├── toDouble
    │   └── toInteger
    ├── Optional
    │   ├── all
    │   ├── any
    │   ├── build
    │   ├── concat
    │   ├── default
    │   ├── filter
    │   ├── fold
    │   ├── head
    │   ├── last
    │   ├── length
    │   ├── map
    │   ├── null
    │   ├── package.dhall
    │   ├── toList
    │   └── unzip
    ├── Text
    │   ├── concat
    │   ├── concatMap
    │   ├── concatMapSep
    │   ├── concatSep
    │   ├── default
    │   ├── defaultMap
    │   ├── package.dhall
    │   └── show
    ├── XML
    │   ├── Type
    │   ├── attribute
    │   ├── element
    │   ├── emptyAttributes
    │   ├── leaf
    │   ├── package.dhall
    │   ├── render
    │   └── text
    └── package.dhall

    Browse the Prelude online to learn more by seeing what functions are available and reading their inline documentation:

    https://prelude.dhall-lang.org

    Exercise: Try to use a new Prelude function that has not been covered previously in this tutorial

    You can also import the entire Prelude as a single large record for convenience:

    $ dhall
        let Prelude = https://prelude.dhall-lang.org/package.dhall
    
    in    λ(x : Text)
        → Prelude.List.length Text (Prelude.List.replicate 10 Text x)
    <Ctrl-D>
    ∀(x : Text) → Natural
    
    λ(x : Text) → 10

    The organization of the package mirrors the layout of the Prelude, meaning that every directory is stored as a record whose children are the fields of that record.

    Exercise: Browse the Prelude by running:

    $ dhall <<< 'https://prelude.dhall-lang.org/package.dhall'

    Limitations

    FromDhall has a limitiation: It won't work for recursive types. If you instantiate these using their Generic instance you end up with one of the following two cases:

    If the type appears in it's own definition like

    data Foo = Foo Foo
        deriving Generic
    
    instance FromDhall Foo

    the resulting Decoder will throw an exception ON USAGE. However, if the recursion is indirect like

    data Foo = Foo Bar
        deriving Generic
    
    instance FromDhall Foo
    
    data Bar = Bar Foo
        deriving Generic
    
    instance FromDhall Bar

    the resulting Decoder will NOT TERMINATE ON USAGE.

    Conclusion

    By this point you should be able to use the Dhall configuration language to author, import, and program configuration files. If you run into any issues you can report them at:

    https://github.com/dhall-lang/dhall-haskell/issues

    You can also request features, support, or documentation improvements on the above issue tracker.

    If you would like to contribute to the Dhall project you can try to port Dhall to languages that do not yet have Dhall integrations so that Dhall configuration files can be read into those languages, too.

    Frequently Asked Questions (FAQ)