Hask

Hask is a pure-Python, zero-dependencies library that mimics most of the core language tools from Haskell, including:

• Full Hindley-Milner type system (with typeclasses) that will typecheck any function decorated with a Hask type signature
• Easy creation of new algebraic data types and new typeclasses, with Haskell-like syntax
• Pattern matching with case expressions
• Automagical function currying/partial application and function composition
• Efficient, immutable, lazily evaluated List type with Haskell-style list comprehensions
• All your favorite syntax and control flow tools, including operator sections, monadic error handling, guards, and more
• Python port of (some of) the standard libraries from Haskell's base, including:
  • Algebraic datatypes from the Haskell Prelude, including Maybe and Either
  • Typeclasses from the Haskell base libraries, including Functor, Applicative, Monad, Enum, Num, and all the rest
  • Standard library functions from base, including all functions from Prelude, Data.List, Data.Maybe, and more

Features not yet implemented, but coming soon:

• Python 3 compatibility
• Better support for polymorphic return values/type defaulting
• Better support for lazy evaluation (beyond just the List type and pattern matching)
• More of the Haskell standard library (Control.* libraries, QuickCheck, and more)
• Monadic, lazy I/O

Note that all of this is still very much pre-alpha, and some things may be buggy!

Installation

1. git clone https://github.com/billpmurphy/hask

2. python setup.py install

To run the tests: python tests.py.

Why did you make this?

I wanted to cram as much of Haskell into Python as possible while still being 100% compatible with the rest of Python, just to see if any useful ideas came out of the result. Also, it was fun!

Contributions, forks, and extensions to this experiment are always welcome! Feel free to submit a pull request, open an issue, or email me. In the spirit of this project, abusing the Python language as much as possible is encouraged.

Features

Hask is a grab-bag of features that add up to one big pseudo-Haskell functional programming library. The rest of this README lays out the basics.

I recommend playing around in the REPL while going through the examples. You

To import all the language features: from hask import * To import the Prelude: from hask import Prelude To import a base library, e.g. Data.List: from hask import Data.List

The List type and list comprehensions

Hask provides the List type, a lazy and statically-typed list, similar to Haskell's standard list type.

To create a new List, just put the elements inside L[ and ] brackets, or wrap an existing iterable inside L[ ].

>>> L[1, 2, 3]
L[1, 2, 3]

>>> my_list = ["a", "b", "c"]
>>> L[my_list]
L['a', 'b', 'c']

>>> L[(x**2 for x in range(1, 11))]
L[1 ... ]

To add elements to the front of a List, use ^, the cons operator. To combine two lists, use +, the concatenation operator.

>>> 1 ^ L[2, 3]
L[1, 2, 3]

>>> "goodnight" ^ ("sweet" ^ ("prince" ^ L[[]]))
L["goodnight", "sweet", "prince"]

>>> "a" ^ L[1.0, 10.3]  # type error

>>> L[1, 2] + L[3, 4]
L[1, 2, 3, 4]

Lists are always evaluated lazily, and will only evaluate list elements as needed, so you can use infinite Lists or put never-ending generators inside of a List. (Of course, you can still blow up the interpreter if you try to evaluate the entirety of an infinite List, e.g. by trying to find the length of the List with len.)

One way to create infinite lists is via list comprehensions. As in Haskell, there are four basic type of list comprehensions:

# list from 1 to infinity, counting by ones
L[1, ...]

# list from 1 to infinity, counting by twos
L[1, 3, ...]

# list from 1 to 20 (inclusive), counting by ones
L[1, ..., 20]

# list from 1 to 20 (inclusive), counting by fours
L[1, 5, ..., 20]

List comprehensions can be used on ints, longs, floats, one-character strings, or any other instance of the Enum typeclass (more on this later).

Hask provides all of the Haskell functions for List manipulation (take, drop, takeWhile, etc.), or you can also use Python-style indexing.

>>> L[1, ...]
L[1 ...]


>>> from hask.Data.List import take
>>> take(5, L["a", "b", ...])
L['a', 'b', 'c', 'd', 'e']


>>> L[1,...][5:10]
L[6, 7, 8, 9, 10]


>>> from hask.Data.List import map
>>> from hask.Data.Char import chr
>>> letters = map(chr, L[97, ...])
>>> letters[:9]
L['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i']


>>> len(L[1, 3, ...])  # uh oh

Otherwise, you can use List just like you would use a regular Python list.

for i in L[0, ...]:
    print i


>>> 55 in L[1, 3, ...]
True

Algebraic Data Types

Hask allows you to define algebraic datatypes, which are immutable objects with a fixed number of typed, unnamed fields.

Here is the definition for the infamous Maybe type:

from hask import data, d, deriving
from hask import Read, Show, Eq, Ord

Maybe, Nothing, Just =\
    data.Maybe("a") == d.Nothing | d.Just("a") & deriving(Read, Show, Eq, Ord)

Let's break this down a bit. The syntax for defining a new type constructor is:

data.TypeName("type param", "type param 2" ... "type param n")

This defines a new algebraic datatype with type parameters.

To define data constructors for this type, use d. The name of the data constructor goes first, followed by its fields. Multiple data constructors should be separted by |. If your data constructor has no fields, you can omit the parens. For example:

FooBar, Foo, Bar =\
    data.FooBar("a", "b") == d.Foo("a", "b", str) | d.Bar

To automagically derive typeclass instances for the type, add & deriving(...typeclasses...) after the data constructor declarations. Currently, the only typeclasses that can be derived are Eq, Show, Read, Ord, and Bounded.

Putting it all together, here are the definitions of Either and Ordering:

Either, Left, Right =\
    data.Either("a", "b") == d.Left("a") | d.Right("b") & deriving(Read, Show, Eq)


Ordering, LT, EQ, GT =\
    data.Ordering == d.LT | d.EQ | d.GT & deriving(Read, Show, Eq, Ord, Bounded)

You can now use the data constructors defined in a data statement to create instances of these new types. If the data constructor takes no arguments, you can use it just like a variable.

>>> Just(10)
Just(10)

>>> Nothing
Nothing

>>> Just(Just(10))
Just(Just(10))

>>> Left(1)
Left(1)

>>> Foo(1, 2, "hello")
Foo(1, 2, 'hello')

You can view the type of an object with _t (equivalent to :t in ghci).

>>> from hask import _t

>>> _t(1)
int

>>> _t(Just("soylent green"))
(Maybe str)

>>> _t(Right(("a", 1)))
(Either a (str, int))

>>> _t(Just)
(a -> Maybe a)

>>> _t(L[1, 2, 3, 4])
[int]

The type system and typed functions

So what's up with those types? Hask operates its own shadow Hindley-Milner type system on top of Python's type system; _t shows the Hask type of a particular object.

In Hask, typed functions take the form of TypedFunc objects, which are typed wrappers around Python functions. There are two ways to create TypedFunc objects:

1. Use the sig decorator to decorate the function with the type signature

@sig(H/ "a" >> "b" >> "a")
def const(x, y):
    return x

2. Use the ** operator (similar to :: in Haskell) to provide the type. Useful for turning functions or lambdas into TypedFunc objects in the REPL, or wrapping already-defined Python functions.

def const(x, y):
    return x

const = const ** (H/ "a" >> "b" >> "a")

TypedFunc objects have several special properties. First, they are type checked--when arguments are supplied, the type inference engine will check whether their types match the type signature, and raise a TypeError if there is a discrepancy.

>>> f = (lambda x, y: x + y) ** (H/ int >> int >> int)

>>> f(2, 3)
5

>>> f(9, 1.0)  # type error

Second, TypedFunc objects can be partially applied:

>>> g = (lambda a, b, c: a / (b + c)) ** (H/ int >> int >> int >> int)

>>> g(10, 2, 3)
2

>>> part_g = g(12)

>>> part_g(2, 2)
3

>>> g(20, 1)(4)
4

TypedFunc objects also have two special infix operators, the * and % operators. * is the compose operator (equivalent to (.) in Haskell), so f * g is equivalent to lambda x: f(g(x)). % is just the apply operator, which applies a TypedFunc to one argument (equivalent to ($) in Haskell). The convinience of this notation (when combined with partial application) cannot be overstated--you can get rid of a ton of nested parenthesis this way.

>>> from hask.Prelude import flip
>>> h = (lambda x, y: x / y) ** (H/ float >> float >> float)

>>> h(3.0) * h(6.0) * flip(h, 2.0) % 36.0
9.0

The compose operation is also typed-checked, which makes it appealing to write programs in pointfree style, i.e, chaining together lots of functions with composition and relying on the type system to catch programming errors.

As you would expect, data constructors are also just TypedFunc objects:

>>> Just * Just * Just * Just % 77
Just(Just(Just(Just(77))))

The type signature syntax is very simple,