from dataclasses import dataclass from functools import partial, wraps from operator import not_ from typing import Any, Callable, Concatenate, Generic, Iterable, Iterator, List, ParamSpec, Sequence, Tuple, TypeVar A = TypeVar('A') B = TypeVar('B') C = TypeVar('C') D = TypeVar('D') P = ParamSpec('P') P1 = ParamSpec('P1') P2 = ParamSpec('P2') # Compose def c(f2: Callable[[B], C], f1: Callable[P, B]) -> Callable[P, C]: """ Compose two functions by passing the output of the second to the input of the first. `c(f1, f2)(*args)` is equivalent to `f1(f2(*args))`. This can also be thought of as mapping the output of a function using the first parameter as a mapper function. >>> double = lambda x: x + x >>> succ = lambda x: x + 1 >>> c(double, succ)(1) 4 >>> c(succ, double)(1) 3 """ @wraps(f1) def inner(*args: P.args, **kwargs: P.kwargs) -> C: return f2(f1(*args, **kwargs)) return inner # Flip: (A -> B -> C) -> B -> A -> C def flip(f: Callable[P1, Callable[P2, C]]) -> Callable[P2, Callable[P1, C]]: """ Reverse the order of the first two arguments of a curried function. This only works with curried functions, so apply `cur2` or `cur3` before applying `flip` if the arguments you want to flip are not curried. >>> pair = lambda x: lambda y: (x, y) >>> pair(1)(2) (1, 2) >>> flip(pair)(1)(2) (2, 1) """ @wraps(f) def inner1(*args2: P2.args, **kwargs2: P2.kwargs) -> Callable[P1, C]: @wraps(f) def inner2(*args1: P1.args, **kwargs1: P1.kwargs) -> C: return f(*args1, **kwargs1)(*args2, **kwargs2) return inner2 return inner1 # Identity function! def ident(x: A) -> A: """ The identity function. Output is identical to input. >>> ident(3) 3 >>> ident(('hello', 8)) ('hello', 8) """ return x def replace(replace_with: A) -> Callable[..., A]: """ Get a function which always returns a constant value, regardless of input The argument `replace_with` is the value the the returned function should always return. The returned function can be used as if having any arity, and will always return the same value originally passed to `replace`. >>> always_seven = replace(7) >>> always_seven(2) 7 >>> always_seven('hello', 'world!') 7 >>> replace('uwu')('NYA!') 'uwu' """ def constant(*args: Any, **kwargs: Any) -> A: "Always return a constant value, typically the one passed to `replace`" return replace_with return constant # Partial Appliaction shorthand p = partial "An alias for partial application" # Two and three-argument currying # Defining these pointfree fucks up the types btw def cur2(f: Callable[Concatenate[A, P], C]) -> Callable[[A], Callable[P, C]]: """ Perform two-argument currying. For example, a function from (A, B) -> C becomes a function A -> B -> C. This can also be though of as simply moving the first argument of a function out front, since it preserves any arguments after the first. That is, a function (A, B, C, kw=D) -> E becomes the function A -> (B, C, kw=D) -> E after being curried using this function. Can also be used as an annotation. >>> @cur2 ... def pair(x, y): ... return (x, y) ... >>> pair(1)(2) (1, 2) >>> alternate_pair = lambda x, y: (x, y) >>> cur2(alternate_pair)(1)(2) (1, 2) >>> threeple = lambda x, y, z: (x, y, z) >>> cur2(threeple)(1)(2, 3) (1, 2, 3) """ return p(p, f) #type:ignore def cur3(f: Callable[Concatenate[A, B, P], D]) -> Callable[[A], Callable[[B], Callable[P, D]]]: """ Perform three-argument currying. See `cur2` for an explaination of how this works. >>> threeple = lambda x, y, z: (x, y, z) >>> cur3(threeple)(1)(2)(3) (1, 2, 3) """ return p(p, p, f) #type:ignore # Curried versions of map & filter with stricter types def p_map(f: Callable[[A], B]) -> Callable[[Sequence[A]], Sequence[B]]: "A curried version of the built in `map` function" return partial(map, f) #type: ignore def p_filter(f: Callable[[A], bool]) -> Callable[[Sequence[A]], Sequence[A]]: "A curried version of the built in `filter` function" return partial(filter,f) #type: ignore # Normal Accessors @cur2 def indx(i: int, s: Sequence[A]) -> A: """ A curried version of the getitem function >>> get_second = indx(1) >>> get_second(('a', 'b')) 'b' >>> get_second([1, 2, 3, 4]) 2 """ return s[i] fst = indx(0) "Get the first element of a tuple/sequence" snd = indx(1) "Get the second element of a tuple/sequence" # Semantic Editor Combinators class SemEdComb: """ A tool which approximates semantic editor combinators in python. Please read https://web.archive.org/web/20221202200001/http://conal.net/blog/posts/semantic-editor-combinators for context. Since Python has no infix function composition, using this pattern can get pretty ugly. This class abuses python's ability to override the property accessor (.) in order to approximate semantic editor combinators. >>> my_func = lambda x: ('abc' + x, 'def') >>> my_func('hi') ('abchi', 'def') >>> altered_func = result.first.map(str.upper, my_func) >>> altered_func('hi') ('ABCHI', 'def') >>> other_altered_func = arg.map(str.upper, my_func) >>> other_altered_func('hello') ('abcHELLO', 'def') Unfortunately, due to limitations of Python's type system, this class is largely untyped. """ class Inner(): "A chain of semantic editor combinators already paired with a map function" def __init__(self, f: Callable, name: str): self.f = f self.name = name def and_then(self, other: 'SemEdComb.Inner') -> 'SemEdComb.Inner': "Composes this with another `SemEdComb.Inner`" return SemEdComb.Inner(c(other.f, self.f), self.name + ' and ' + other.name) def __repr__(self) -> str: return f"SemEdComb*({self.name})" def __call__(self, *args, **kwargs): return self.f(*args, **kwargs) def __init__(self, f: Callable[[Callable],Callable], name: str): self.f = f self.name = name def _c(self, next_f: Callable[[Callable], Callable], next_fname: str) -> 'SemEdComb': return SemEdComb(c(self.f, next_f), self.name + next_fname) RESULT = cur2(c) "Map the result of a function" ARG = flip(RESULT) "Map the argument of a function" ALL = p_map "Map every element of a list" @cur3 @staticmethod def INDEX(i, f, arr): "Map the ith element of a mutable sequence" arr[i] = f(arr[i]) return arr @cur3 @staticmethod def INDEX_TUP(i: int, f: Callable[[Any], Any], tup: Tuple) -> Tuple: "Map the ith element of an immutable sequence" l = list(tup) l[i] = f(l[i]) return (*l,) @cur2 @staticmethod def FIRST(f: Callable[[A], C], tup: Tuple[A, B]) -> Tuple[C, B]: "Map the first element of a two-tuple" return (f(tup[0]), tup[1]) @cur2 @staticmethod def SECOND(f: Callable[[B], C], tup: Tuple[A, B]) -> Tuple[A, C]: "Map the second element of a two-tuple" return (tup[0], f(tup[1])) @property def result(self) -> 'SemEdComb': """ Map the result of a function >>> my_func = lambda s: s + ' backwards is ' + s[::-1] >>> my_func('hello') 'hello backwards is olleh' >>> altered_func = result.map(str.upper, my_func) >>> altered_func('hello') 'HELLO BACKWARDS IS OLLEH' Can be chained in order to work with curried functions as well. That is, the result of a two argument curried function is the result of the result of that function. >>> curried_pair = lambda x: lambda y: (x, y) >>> altered_pair = result.result.second.map(str.upper, curried_pair) >>> altered_pair('hello')('world') ('hello', 'WORLD') """ return self._c(SemEdComb.RESULT, '.result') @property def arg(self) -> 'SemEdComb': """ Map the argument of a function >>> my_func = lambda s: s + ' backwards is ' + s[::-1] >>> my_func('hello') 'hello backwards is olleh' >>> altered_func = arg.map(str.upper, my_func) >>> altered_func('hello') 'HELLO backwards is OLLEH' Can be combined with `.result` to work with curried functions. >>> curried_pair = lambda x: lambda y: (x, y) >>> altered_pair = result.arg.map(str.upper, curried_pair) >>> altered_pair('hello')('world') ('hello', 'WORLD') """ return self._c(SemEdComb.ARG, '.arg') @property def all(self) -> 'SemEdComb': """ Map every element of a sequence To use this as the base of a chain of SECs, write "all_", since "all" by itself refers to the builtin python function, which is different. Note that this returns an iterator, not a sequence, even if the thing being mapped was a sequence or a list. >>> list(all_.map(lambda x: x + x, [1, 2, 3])) [2, 4, 6] >>> my_func = lambda s: [s] * s >>> my_func(3) [3, 3, 3] >>> altered_func = result.all.map(lambda x: x + x, my_func) >>> list(altered_func(3)) [6, 6, 6] """ return self._c(SemEdComb.ALL, '.all') def index(self, i) -> 'SemEdComb': """ Map the ith element of a mutable sequence >>> index(1).map(lambda x: x + x, [1, 2, 3]) [1, 4, 3] >>> my_func = lambda s: [s] * s >>> my_func(3) [3, 3, 3] >>> altered_func = result.index(1).map(lambda x: x + x, my_func) >>> list(altered_func(3)) [3, 6, 3] """ return self._c(SemEdComb.INDEX(i), f'.index({i})') def index_tup(self, i) -> 'SemEdComb': """ Map the ith element of an immutable sequence. >>> index_tup(2).map(lambda x: x + x, (1, 2, 3, 4)) (1, 2, 6, 4) See Also: `index` For a more optimized version of this method specialized to two-tuples, see `first` and `second` """ return self._c(SemEdComb.INDEX_TUP(i), f'.index_tup({i})') @property def first(self) -> 'SemEdComb': """ Map the first element of a two-tuple >>> first.map(lambda x: x+x, (1, 2)) (2, 2) Doesn't work for threeples and fourples. If this is the behaviour you need, try `index_tup` >>> first.map(lambda x: x+x, (1, 2, 3)) (2, 2) """ return self._c(SemEdComb.FIRST, f'.first') @property def second(self) -> 'SemEdComb': """ Map the second element of a two-tuple >>> second.map(lambda x: x+x, (1, 2)) (1, 4) As with `first`, this doesn't work with threeples, fourples, and moreples. >>> second.map(lambda x: x+x, (1, 2, 3)) (1, 4) """ return self._c(SemEdComb.SECOND, f'.second') def __repr__(self): return f"SemEdComb({self.name})" def pmap(self, mapper): """ Set the mapper function, but don't call it yet The name is short for partial map. >>> my_func = lambda s1: lambda s2: f"You entered {s1} and the pair {s2}" >>> my_func(1)(('hello', 'world')) "You entered 1 and the pair ('hello', 'world')" >>> mapper = result.arg.first.pmap(str.upper) >>> altered_func = mapper(my_func) >>> altered_func(1)(('hello', 'world')) "You entered 1 and the pair ('HELLO', 'world')" See also: `map` """ return SemEdComb.Inner(self.f(mapper), self.name) def map(self, mapper, thing_to_map) -> Callable: "Apply the chain of combinators to a mapper and a mappee" return self.pmap(mapper)(thing_to_map) def __call__(self, *args, **kwargs): return self.f(*args, **kwargs) # Pre-constructed base semantic editor combinators result = SemEdComb(SemEdComb.RESULT, 'result') arg = SemEdComb(SemEdComb.ARG, 'arg') index = lambda i: SemEdComb(SemEdComb.INDEX(i), f'index({i})') index_tup = lambda i: SemEdComb(SemEdComb.INDEX_TUP(i), f'index_tup({i})') first = SemEdComb(SemEdComb.FIRST, 'first') second = SemEdComb(SemEdComb.SECOND, 'second') all_ = SemEdComb(SemEdComb.ALL, 'all') # Tail call optimizing recursion @dataclass class Recur(Generic[P]): """ Indicate that the function this is returned from should be called again with new args. Exclusively used with `tco_rec()` """ def __init__(self, *args: P.args, **kwargs: P.kwargs): self.args = args self.kwargs = kwargs @dataclass(frozen = True) class Return(Generic[B]): """ Indicate that the function this is returned from should return this value Exclusively used with `tco_rec()` """ val: B def tco_rec(f: Callable[P, Recur[P] | Return[B]]) -> Callable[P, B]: """ Run a tail-recursive function in a mannor which will not overflow the stack. Wraps a function in a loop which transforms its return type. The function is expected to return an instance of `Recur` rather than calling itself to recur. The arguments passed to the returned `Recur` instance become the arguments to the next iteration of the function call. When the function is ready to return for real, it should return an instance of `Return`. The function will be transformed by `tco_rec` to look as if it is a normal function. >>> @tco_rec ... def factorial(n, coefficient = 1): ... if n > 1: ... return Recur(n - 1, coefficient * n) ... else: ... return Return(coefficient) >>> factorial(4) 24 """ @wraps(f) def tco_loop(*args: P.args, **kwargs: P.kwargs) -> B: while True: match f(*args, **kwargs): case Recur(args=args, kwargs=kwargs): #type:ignore pass case Return(val=val)|val: return val #type:ignore return tco_loop # Options! @dataclass(frozen=True) class Some(Generic[A]): """ The positive part of an optional datatype Component of `Option` and counterpart of `None` """ val: A def __repr__(self) -> str: return f'Some({self.val!r})' Option = Some[A] | None "An Option datatype, aka Maybe" def map_opt(f: Callable[[A], B], o: Option[A]) -> Option[B]: """ Map the contents of an optional data type. Has no effect on `None` >>> map_opt(str.upper, Some('hello')) Some('HELLO') >>> map_opt(str.upper, None) is None True """ match o: case Some(val): return Some(f(val)) case none: return none def bind_opt(f: Callable[[A], Option[B]], o: Option[A]) -> Option[B]: """ wow! monads! (aka 'and_then') >>> halve = lambda n: Some(n//2) if n % 2 == 0 else None >>> [halve(2), halve(3)] [Some(1), None] >>> bind_opt(halve, Some(4)) Some(2) >>> bind_opt(halve, Some(5)) is None True >>> bind_opt(halve, None) is None True """ match o: case Some(val): return f(val) case none: return none def note(e: B, o: Option[A]) -> 'Result[A, B]': """ Convert an `Option` to a `Result` by attaching an error to the `None` variants >>> note('woops!', Some(1)) Ok(1) >>> note('woops!', None) Err('woops!') """ match o: case Some(val): return Ok(val) case None: return Err(e) # Results! @dataclass(frozen=True) class Ok(Generic[A]): """ The positive part of a result (either) datatype Component of `Result` and counterpart of `Err` """ val: A def __repr__(self) -> str: return f'Ok({self.val!r})' @dataclass(frozen=True) class Err(Generic[B]): """ The error part of a result (either) datatype Component of `Result` and counterpart of `Ok` """ err: B def __repr__(self) -> str: return f'Err({self.err!r})' def __bool__(self): return False Result = Ok[A] | Err[B] "A Result datatype, aka Either" def map_res(f: Callable[[A], C], r: Result[A, B]) -> Result[C, B]: """ Map the success value of a result >>> map_res(str.upper, Ok('hai!')) Ok('HAI!') >>> map_res(str.upper, Err('oh noes')) Err('oh noes') """ match r: case Ok(val): return Ok(f(val)) case not_okay: return not_okay def bind_res(f: Callable[[A], Result[C, B]], r: Result[A, B]) -> Result[C, B]: """ Perform an fallible operation for successful results. >>> halve = lambda n: Ok(n//2) if n % 2 == 0 else Err(f'{n} is not divisible by 2') >>> [halve(2), halve(3)] [Ok(1), Err('3 is not divisible by 2')] >>> bind_res(halve, Ok(4)) Ok(2) >>> bind_res(halve, Ok(5)) Err('5 is not divisible by 2') >>> bind_res(halve, Err('not okay in the 1st place')) Err('not okay in the 1st place') """ match r: case Ok(val): return f(val) case not_okay: return not_okay def map_err(f: Callable[[B], C], r: Result[A, B]) -> Result[A, C]: """ Map the error value of a result >>> map_err(str.upper, Ok('hai!')) Ok('hai!') >>> map_err(str.upper, Err('oh noes')) Err('OH NOES') """ match r: case Err(e): return Err(f(e)) case oki_doke: return oki_doke def hush(r: Result[A, Any]) -> Option[A]: """ Convert a `Result` to an `Option` by converting any errors to `None` >>> hush(Ok('hai!')) Some('hai!') >>> hush(Err('oh noes')) is None True """ match r: case Ok(val): return Some(val) case not_okay: return None def try_(handle: Callable[[Exception], B], f: Callable[P, A], *args: P.args, **kwargs: P.kwargs) -> Result[A, B]: """ Try-catch in a function! Attempt to perform and operation, and `Err` on failure Arguments: handle - A function which handles any exceptions which arise. The return type is what will be wrapped into the resulting `Err`. This is not called if nothing goes wrong. f - The fallible function to try. If this succeeds without raising an error, that value is returned in an `Ok`. If this raises an exception, that exception will be passed to `handle`. args - Will be passed to `f` when it is called. kwargs - Will be passed to `f` when it is called. >>> try_(ident, int, '3') Ok(3) >>> try_(ident, int, 'three') Err(ValueError("invalid literal for int() with base 10: 'three'")) """ try: return Ok(f(*args, **kwargs)) except Exception as e: return Err(handle(e)) def unwrap_r(r: Result[A, Any]) -> A: """ Assert that a `Result` is `Ok` and return it's value. Throws: `AssertionError` - The result was NOT okay. The `AssertionError` will have two arguments: The first is a string to make it more obvious what happened. The second is the error that was stored in the `Err`. >>> unwrap_r(Ok('hai!')) 'hai!' >>> unwrap_r(Err('oh noes')) is None #doctest: +IGNORE_EXCEPTION_DETAIL Traceback (most recent call last): AssertionError: ('Tried to unwrap an error: ', 'oh noes') """ match r: case Ok(val): return val case Err(e): raise AssertionError(f'Tried to unwrap an error: ', e) def sequence(s: Sequence[Result[A, B]]) -> Result[Iterator[A], B]: """ Convert a list of results into a result of a list. If the input sequence contains only `Ok` results, then the output is similarly `Ok`, and contains a list of all the unwrapped values of the `Ok`s. If there are any errors, proccessing of the sequence is immediately stopped, and the first error encountered is returned. >>> map_res(list, sequence([Ok(1), Ok(2), Ok(3)])) Ok([1, 2, 3]) >>> sequence([Ok(1), Err('Oops!'), Err('Aw man!')]) Err('Oops!') """ if all(s): return Ok(map(unwrap_r, s)) else: o = next(filter(not_, s)) assert isinstance(o, Err) return o if __name__ == '__main__': import doctest doctest.testmod()