The Type System¶
Like many functional-style programming languages, the power of Hobbes lies in its rich type system, so that’s where we’ll start. There are primitive types, arrays, record types, tuples, and variants.
Hint
hi, the Hobbes interpreter
You can execute most of the code examples we’ll show here in hi, the interpreter that comes packaged with Hobbes. For more information, take a look at the chapter hi, the Hobbes interpreter
Primitive Types¶
The Hobbes primitive types are arranged in memory in a manner which allows for free marshalling when Hobbes is executing in a C++ process. Each of the primitive types has a simple literal syntax allowing for the easy initialisation of a value:
- Unit
A simple null-type with only one value
> ()
- Bool
Either
trueorfalse> true true
- Char
A single character of text
> 's' s
- Byte
A single byte (
0-255)> 0Xad 0Xad
- Short
A two-byte number
> 3S 3S
- Int
A four-byte number
> 4 4
- Long
An eight-byte number
> 5L 5
- Float
A single-precision (4 byte) floating-point number
> 3.0f 3f
Note
float literals must be a decimal number followed by the character ‘f’.
- Double
A double-precision (8-byte) floating-point number
> 3.0 3
Arrays¶
Just like in C++, Hobbes arrays are zero-indexed, contiguous values in memory, with special syntax for char and byte arrays. Hobbes supports bounds-checking to prevent a common class of bug by maintaining the array length alongside the data in memory.
> [0, 1, 1, 2, 3]
[0, 1, 1, 2, 3]
> "Hello, Hobbes!"
"Hello, Hobbes!"
> 0xdeadbeef
0xdeadbeef
Note
Strings
In Hobbes, a String is simply an array of char.
Warning
0x versus 0X
It’s important the note the subtle difference between the literal syntax for byte and for byte arrays - the case of the ‘X’ is very important!
Uppercase for byte, lowercase for byte array.
Array functions¶
A number of functions are overloaded for array types:
> [0, 1, 2] ++ [3, 4, 5]
[0, 1, 2, 3, 4, 5]
> size([0,1,2])
3
You can index into an array using square brackets:
> nums = [6, 2, 4, 6, 5, 9, 8, 5, 6, 3]
> nums[3]
6
In addition, the open and closed slice syntax is available:
> nums[3:6]
[6, 5, 9]
> nums[2:]
[4, 6, 5, 9, 8, 5, 6, 3]
You can read [2:] as “the second index, until the end”. The converse works, too:
> nums[:2]
[3, 6, 5, 8, 9, 5, 6, 4]
This is “the end until the second index”. Indexes from the end of the array can be counted with a unary negate:
> nums[2:-3]
[4, 6, 5, 9, 8]
And of course you do that in both positions in the slice:
> nums[-2:-4]
[5, 8]
Warning
Array indexes
Array indexes in Hobbes aren’t bounds checked, so whilst you can slice from the end of an array, you can’t use the same syntax to index:
> nums[-3]
51627831
In addition, attempts to slice off the end of an array will act as though you were slicing from the beginning or the end, respectively:
> nums[-2014:305]
[6, 2, 4, 6, 5, 9, 8, 5, 6, 3]
Array sequences¶
A sequence expression can be used to initialise an array of ints. The syntax is simple:
> [1..4]
[1, 2, 3, 4]
We can take this further and generate infinite sequences by leaving the upper bound open:
> [0..]
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]...
Infinite sequences are useful structures for performing work many times over without mutating a loop variable. Special care has been taken to ensure their evaluation isn’t eager, however - as your program might never stop! For more information about the type of an infinite sequence, take a look at the infinite squences section in polymorphism
Records¶
Records are a common way to keep closely-associated pieces of data together in functional progamming, and they’re often referred to as an and type: a hostport is a host and a port - and that’s it. No behaviour, and its identity is simply the two elements.
Record types are similar in spirit to C++ structs, with ad-hoc declaration and initialisation, plus type inference:
> {name="Sam", age=23, job="writer"}
Records are examples of structural types, meaning that in Hobbes, even though they are both examples of different anonymous ad-hoc types, the two are equivalent:
> {name="Sam", job="Writer"} == {job="Writer", name="Sam"}
true
Note
Equivalence vs Equality Although it’s true to say that, in Hobbes, the two record instances above are equivalent, they’re not equal, and so the following equality test would fail to compile:
> {name="Sam", job="Writer"} === {job="Writer", name="Sam"}
stdin:1,28-30: Cannot unify types: { name:[char], job:[char] } != { job:[char], name:[char] }
This is because the equivalence relationship is determined not by any special logic in the Hobbes compiler, but by the equivalency type class Equiv. This class contains the implementation of == and thus decides how to unpack the record instances and compare them.
A type class is a way of describing expected behaviour on a type. In the Hi REPL, I can unpack the Equiv typeclass with :c:
> :c Equiv
class Equiv where
== :: (#0 * #1) -> bool
For more information about typeclasses in Hobbes, see Type Classes.
Tuples¶
Like records but with no field names, tuples are used to keep commonly-associated data together. The canonical example is the host/port pair:
> endpoint = ("lndev1", 3923)
> endpoint
("lndev1", 3923)
Note
Assignment
Notice here that we’ve assigned the tuple to the name endpoint. This name now exists in the global context, which means you can refer to those values using the name “endpoint” from now on. For local scoping, see Local scoping
Note
Pretty-printing
Hobbes has good support for printing the primitive and scalar types: char arrays are printed as strings, the literal syntax is displayed when printing to standard out, etc.
When we deal with arrays of records or tuples, Hobbes gives us a convenient table notation:
> [{First=1, Second="two"},{First=3, Second="Four"},{First=5, Second="Six"}]
First Second
----- ------
1 two
3 Four
5 Six
Variants¶
The variant is the richest way to declare a type in the Hobbes type system, because it gives us the opportunity to declare a value which can be one of a number of named cases. If the Record type is an and, the Variant is an or.
This allows us to model enum-like structures with associated data. In the following example, we’re declaring a type called status which can model the success or failure of a service call. In the case of a failure, we’ll be given an error code which we’ll want to react to. However, in the successful case, there’s nothing more to do:
type status = |
Succeeded,
Failed: int
|
> status = |Succeeded| :: status
|Succeeded|
Warning
type declarations in hi
Hi doesn’t currently support some Hobbes expressions, including type declarations. You can write your types in a file and have them loaded into a hi session by following the instructions in Hi can load files.
Note
Type Annotations
Sometimes Hobbes requires us to specify the type of a value. In the case above, we want to be careful about the instantiation of the |Succeeded| type: we need to be clear that we’re instantiating a subtype of status, rather than a naked record type with just one subtype which happens to be called ‘Succeeded’. hi can show us the inferred type of a value with :t:
> :t |Succeeded|
|Succeeded=()|::a=>a
> :t |Succeeded| :: status
|Succeeded, Failed:int|
The :: allows us to specify the type of the variable using what’s called a type annotation. More information about types and type annotations is available in Polymorphism in Hobbes.
As we’ll see in pattern matching, Hobbes has rich language support for building logic based on variant types.
Sum types¶
Just as the tuple type can be thought of as simply a record using numbered placement instead of names, the sum is a variant without names: a true union.
> |1="hello"| :: (int+[char])
|1="hello"|
> |0=3| :: (int+[char])
|0=3|
In this case we’re using the index (0 or 1) to specify the actual variant type we’re using - int or char array. An instance of the first type must hold an int, and an instance of the second type must hold a char array - in this case, a String.
Recursive type definitions¶
Recursive types (i.e. a type whos definition includes itself) are a powerful way to define complex data structures and are commonly found in functional programming.
The canonical example is the list, which is defined as two parts: head and tail. The head is a single list element, whilst the tail is… another list!
In Hobbes, we declare that a type is recursive by simply giving it a name and denoting the name with a caret:
^x.(()+([char]*x))
In this type expression we use the caret to give a name to the type so it can be used recursively throughout the expression. In this case the list type, x, is declared as a sum type of an empty list, or a string and a list.
We can easily construct one using Hobbes’s constructor syntax:
> cons(1, cons(2, cons(3, nil())))
1:2:3:[]
Whilst this construction syntax might look unwieldy, the generation of such structures is commonly algorhithmic, and (as discussed earlier), the payoff is in Hobbes’s rich matching syntax.
Many structures in Hobbes can be defined recursively because, as we’ll see, recursion is a deeply powerful element of functional programming.