Awesome

Unpythonic: Python meets Lisp and Haskell

In the spirit of toolz, we provide missing features for Python, mainly from the list processing tradition, but with some Haskellisms mixed in. We extend the language with a set of syntactic macros. We also provide an in-process, background REPL server for live inspection and hot-patching. The emphasis is on clear, pythonic syntax, making features work together, and obsessive correctness.

Some hypertext features of this README, such as local links to detailed documentation, and expandable example highlights, are not supported when viewed on PyPI; view on GitHub to have those work properly.

Dependencies

None required.

• mcpyrate optional, to enable the syntactic macro layer, an interactive macro REPL, and some example dialects.

As of v0.15.3, unpythonic runs on CPython 3.8, 3.9 and 3.10, 3.11, 3.12, and PyPy3 (language versions 3.8, 3.9, 3.10); the CI process verifies the tests pass on those platforms. New Python versions are added and old ones are removed following the Long-term support roadmap.

Documentation

• README: you are here.
• Pure-Python feature set
• Syntactic macro feature set
• Examples of creating dialects using mcpyrate: Python the way you want it.
• REPL server: interactively hot-patch your running Python program.
• Troubleshooting: possible solutions to possibly common issues.
• Design notes: for more insight into the design choices of unpythonic.
• Essays: for writings on the philosophy of unpythonic, things that inspired it, and related discoveries.
• Additional reading: links to material relevant in the context of unpythonic.
• Contribution guidelines: for understanding the codebase, or if you're interested in making a code or documentation PR.

The features of unpythonic are built out of, in increasing order of magic:

• Pure Python (e.g. batteries for itertools),
• Macros driving a pure-Python core (do, let),
• Pure macros (e.g. continuations, lazify, dbg).
• Whole-module transformations, a.k.a. dialects (e.g. Lispy).

This depends on the purpose of each feature, as well as ease-of-use considerations. See the design notes for more information.

Examples

Small, limited-space overview of the overall flavor. There is a lot more that does not fit here, especially in the pure-Python feature set. We give here simple examples that are not necessarily of the most general form supported by the constructs. See the full documentation and unit tests for more examples.

Unpythonic in 30 seconds: Pure Python

[docs]

from unpythonic import looped, looped_over

@looped
def result(loop, acc=0, i=0):
    if i == 10:
        return acc
    else:
        return loop(acc + i, i + 1)  # tail call optimized, no call stack blowup.
assert result == 45

@looped_over(range(3), acc=[])
def result(loop, i, acc):
    acc.append(lambda x: i * x)  # fresh "i" each time, no mutation of loop counter.
    return loop()
assert [f(10) for f in result] == [0, 10, 20]

[docs]

from unpythonic import dyn, make_dynvar

make_dynvar(x=42)  # set a default value

def f():
    assert dyn.x == 17
    with dyn.let(x=23):
        assert dyn.x == 23
        g()
    assert dyn.x == 17

def g():
    assert dyn.x == 23

assert dyn.x == 42
with dyn.let(x=17):
    assert dyn.x == 17
    f()
assert dyn.x == 42

[docs]

To opt in, add just two lines of code to your main program:

from unpythonic.net import server
server.start(locals={})  # automatically daemonic

import time

def main():
    while True:
        time.sleep(1)

if __name__ == '__main__':
    main()

Or if you just want to take this for a test run, start the built-in demo app:

python3 -m unpythonic.net.server

Once a server is running, to connect:

python3 -m unpythonic.net.client 127.0.0.1

This gives you a REPL, inside your live process, with all the power of Python. You can importlib.reload any module, and through sys.modules, inspect or overwrite any name at the top level of any module. You can pickle.dump your data. Or do anything you want with/to the live state of your app.

You can have multiple REPL sessions connected simultaneously. When your app exits (for any reason), the server automatically shuts down, closing all connections if any remain. But exiting the client leaves the server running, so you can connect again later - that's the whole point.

Optionally, if you have mcpyrate, the REPL sessions support importing, invoking and defining macros.

[docs]

Scan and fold accept multiple iterables, like in Racket.

from operator import add
from unpythonic import scanl, foldl, unfold, take, Values

assert tuple(scanl(add, 0, range(1, 5))) == (0, 1, 3, 6, 10)

def op(e1, e2, acc):
    return acc + e1 * e2
assert foldl(op, 0, (1, 2), (3, 4)) == 11

def nextfibo(a, b):
    return Values(a, a=b, b=a + b)
assert tuple(take(10, unfold(nextfibo, 1, 1))) == (1, 1, 2, 3, 5, 8, 13, 21, 34, 55)

[docs]

We bind arguments to parameters like Python itself does, so it does not matter whether arguments are passed by position or by name during currying. We support @generic multiple-dispatch functions.

We also feature a Haskell-inspired passthrough system: any args and kwargs that are not accepted by the call signature will be passed through. This is useful when a curried function returns a new function, which is then the target for the passthrough. See the docs for details.

from unpythonic import curry, generic, foldr, composerc, cons, nil, ll

@curry
def f(x, y):
    return x, y

assert f(1, 2) == (1, 2)
assert f(1)(2) == (1, 2)
assert f(1)(y=2) == (1, 2)
assert f(y=2)(x=1) == (1, 2)

@curry
def add3(x, y, z):
    return x + y + z

# actually uses partial application so these work, too
assert add3(1)(2)(3) == 6
assert add3(1, 2)(3) == 6
assert add3(1)(2, 3) == 6
assert add3(1, 2, 3) == 6

@curry
def lispyadd(*args):
    return sum(args)
assert lispyadd() == 0  # no args is a valid arity here

@generic
def g(x: int, y: int):
    return "int"
@generic
def g(x: float, y: float):
    return "float"
@generic
def g(s: str):
    return "str"
g = curry(g)

assert callable(g(1))
assert g(1)(2) == "int"

assert callable(g(1.0))
assert g(1.0)(2.0) == "float"

assert g("cat") == "str"
assert g(s="cat") == "str"

# simple example of passthrough
mymap = lambda f: curry(foldr, composerc(cons, f), nil)
myadd = lambda a, b: a + b
assert curry(mymap, myadd, ll(1, 2, 3), ll(2, 4, 6)) == ll(3, 6, 9)

[docs]

from unpythonic import generic

@generic
def my_range(stop: int):  # create the generic function and the first multimethod
    return my_range(0, 1, stop)
@generic
def my_range(start: int, stop: int):  # further registrations add more multimethods
    return my_range(start, 1, stop)
@generic
def my_range(start: int, step: int, stop: int):
    return start, step, stop

This is a purely run-time implementation, so it does not give performance benefits, but it can make code more readable, and makes it modular to add support for new input types (or different call signatures) to an existing function later.

Holy traits are also a possibility:

import typing
from unpythonic import generic, augment

class FunninessTrait:
    pass
class IsFunny(FunninessTrait):
    pass
class IsNotFunny(FunninessTrait):
    pass

@generic
def funny(x: typing.Any):  # default
    raise NotImplementedError(f"`funny` trait not registered for anything matching {type(x)}")

@augment(funny)
def funny(x: str):  # noqa: F811
    return IsFunny()
@augment(funny)
def funny(x: int):  # noqa: F811
    return IsNotFunny()

@generic
def laugh(x: typing.Any):
    return laugh(funny(x), x)

@augment(laugh)
def laugh(traitvalue: IsFunny, x: typing.Any):
    return f"Ha ha ha, {x} is funny!"
@augment(laugh)
def laugh(traitvalue: IsNotFunny, x: typing.Any):
    return f"{x} is not funny."

assert laugh("that") == "Ha ha ha, that is funny!"
assert laugh(42) == "42 is not funny."

[docs]

Contrived example:

from unpythonic import error, restarts, handlers, invoke, use_value, unbox

class MyError(ValueError):
    def __init__(self, value):  # We want to act on the value, so save it.
        self.value = value

def lowlevel(lst):
    _drop = object()  # gensym/nonce
    out = []
    for k in lst:
        # Provide several different error recovery strategies.
        with restarts(use_value=(lambda x: x),
                      halve=(lambda x: x // 2),
                      drop=(lambda: _drop)) as result:
            if k > 9000:
                error(MyError(k))
            # This is reached when no error occurs.
            # `result` is a box, send k into it.
            result << k
        # Now the result box contains either k,
        # or the return value of one of the restarts.
        r = unbox(result)  # get the value from the box
        if r is not _drop:
            out.append(r)
    return out

def highlevel():
    # Choose which error recovery strategy to use...
    with handlers((MyError, lambda c: use_value(c.value))):
        assert lowlevel([17, 10000, 23, 42]) == [17, 10000, 23, 42]

    # ...on a per-use-site basis...
    with handlers((MyError, lambda c: invoke("halve", c.value))):
        assert lowlevel([17, 10000, 23, 42]) == [17, 5000, 23, 42]

    # ...without changing the low-level code.
    with handlers((MyError, lambda: invoke("drop"))):
        assert lowlevel([17, 10000, 23, 42]) == [17, 23, 42]

highlevel()

Conditions only shine in larger systems, with restarts set up at multiple levels of the call stack; this example is too small to demonstrate that. The single-level case here could be implemented as a error-handling mode parameter for the example's only low-level function.

With multiple levels, it becomes apparent that this mode parameter must be threaded through the API at each level, unless it is stored as a dynamic variable (see unpythonic.dyn). But then, there can be several types of errors, and the error-handling mode parameters - one for each error type - have to be shepherded in an intricate manner. A stack is needed, so that an inner level may temporarily override the handler for a particular error type...

The condition system is the clean, general solution to this problem. It automatically scopes handlers to their dynamic extent, and manages the handler stack automatically. In other words, it dynamically binds error-handling modes (for several types of errors, if desired) in a controlled, easily understood manner. The local programmability (i.e. the fact that a handler is not just a restart name, but an arbitrary function) is a bonus for additional flexibility.

If this sounds a lot like an exception system, that's because conditions are the supercharged sister of exceptions. The condition model cleanly separates mechanism from policy, while otherwise remaining similar to the exception model.

[docs]

Roughly, a symbol is a guaranteed-interned string.

A gensym is a guaranteed-unique string, which is useful as a nonce value. It's similar to the pythonic idiom nonce = object(), but with a nice repr, and object-identity-preserving pickle support.

from unpythonic import sym  # lispy symbol
sandwich = sym("sandwich")
hamburger = sym("sandwich")  # symbol's identity is determined by its name, only
assert hamburger is sandwich

assert str(sandwich) == "sandwich"  # symbols have a nice str()
assert repr(sandwich) == 'sym("sandwich")'  # and eval-able repr()
assert eval(repr(sandwich)) is sandwich

from pickle import dumps, loads
pickled_sandwich = dumps(sandwich)
unpickled_sandwich = loads(pickled_sandwich)
assert unpickled_sandwich is sandwich  # symbols survive a pickle roundtrip

from unpythonic import gensym  # gensym: make new uninterned symbol
tabby = gensym("cat")
scottishfold = gensym("cat")
assert tabby is not scottishfold

pickled_tabby = dumps(tabby)
unpickled_tabby = loads(pickled_tabby)
assert unpickled_tabby is tabby  # also gensyms survive a pickle roundtrip

[docs for box] [docs for cons] [docs for frozendict]

from unpythonic import box, unbox  # mutable single-item container
cat = object()
cardboardbox = box(cat)
assert cardboardbox is not cat  # the box is not the cat
assert unbox(cardboardbox) is cat  # but the cat is inside the box
assert cat in cardboardbox  # ...also syntactically
dog = object()
cardboardbox << dog  # hey, it's my box! (replace contents)
assert unbox(cardboardbox) is dog

from unpythonic import cons, nil, ll, llist  # lispy linked lists
lst = cons(1, cons(2, cons(3, nil)))
assert ll(1, 2, 3) == lst  # make linked list out of elements
assert llist([1, 2, 3]) == lst  # convert iterable to linked list

from unpythonic import frozendict  # immutable dictionary
d1 = frozendict({'a': 1, 'b': 2})
d2 = frozendict(d1, c=3, a=4)
assert d1 == frozendict({'a': 1, 'b': 2})
assert d2 == frozendict({'a': 4, 'b': 2, 'c': 3})

[docs for withself] [docs for namelambda]

from unpythonic import withself, namelambda

fact = withself(lambda self, n: n * self(n - 1) if n > 1 else 1)  # see @trampolined to do this with TCO
assert fact(5) == 120

square = namelambda("square")(lambda x: x**2)
assert square.__name__ == "square"
assert square.__qualname__ == "square"  # or e.g. "somefunc.<locals>.square" if inside a function
assert square.__code__.co_name == "square"  # used by stack traces

[docs]

from typing import NoReturn
from unpythonic import fix

@fix()
def a(k):
    return b((k + 1) % 3)
@fix()
def b(k):
    return a((k + 1) % 3)
assert a(0) is NoReturn

[docs for s] [docs for islice]

from unpythonic import s, islice

seq = s(1, 2, 4, ...)
assert tuple(islice(seq)[:10]) == (1, 2, 4, 8, 16, 32, 64, 128, 256, 512)

[docs for memoize] [docs for gmemoize]

from itertools import count, takewhile
from unpythonic import memoize, gmemoize, islice

ncalls = 0
@memoize  # <-- important part
def square(x):
    global ncalls
    ncalls += 1
    return x**2
assert square(2) == 4
assert ncalls == 1
assert square(3) == 9
assert ncalls == 2
assert square(3) == 9
assert ncalls == 2  # called only once for each unique set of arguments

# "memoize lambda": classic evaluate-at-most-once thunk
thunk = memoize(lambda: print("hi from thunk"))
thunk()  # the message is printed only the first time
thunk()

@gmemoize  # <-- important part
def primes():  # FP sieve of Eratosthenes
    yield 2
    for n in count(start=3, step=2):
        if not any(n % p == 0 for p in takewhile(lambda x: x*x <= n, primes())):
            yield n

assert tuple(islice(primes())[:10]) == (2, 3, 5, 7, 11, 13, 17, 19, 23, 29)

[docs]

from itertools import repeat
from unpythonic import fup

t = (1, 2, 3, 4, 5)
s = fup(t)[0::2] << repeat(10)
assert s == (10, 2, 10, 4, 10)
assert t == (1, 2, 3, 4, 5)

from itertools import count
from unpythonic import imemoize
t = (1, 2, 3, 4, 5)
s = fup(t)[::-2] << imemoize(count(start=10))()
assert s == (12, 2, 11, 4, 10)
assert t == (1, 2, 3, 4, 5)

[docs]

from unpythonic import view

lst = list(range(10))
v = view(lst)[::2]  # [0, 2, 4, 6, 8]
v[2:4] = (10, 20)  # re-slicable, still live.
assert lst == [0, 1, 2, 3, 10, 5, 20, 7, 8, 9]

lst[2] = 42
assert v == [0, 42, 10, 20, 8]

[docs]

from unpythonic import piped, exitpipe

double = lambda x: 2 * x
inc    = lambda x: x + 1
x = piped(42) | double | inc | exitpipe
assert x == 85

The point is usability: in a function composition using pipe syntax, data flows from left to right.

Unpythonic in 30 seconds: Language extensions with macros

[docs]

from unpythonic.syntax import macros, test, test_raises, fail, error, warn, the
from unpythonic.test.fixtures import session, testset, terminate, returns_normally

def f():
    raise RuntimeError("argh!")

def g(a, b):
    return a * b
    fail["this line should be unreachable"]

count = 0
def counter():
    global count
    count += 1
    return count

with session("simple framework demo"):
    with testset():
        test[2 + 2 == 4]
        test_raises[RuntimeError, f()]
        test[returns_normally(g(2, 3))]
        test[g(2, 3) == 6]
        # Use `the[]` (or several) in a `test[]` to declare what you want to inspect if the test fails.
        # Implicit `the[]`: in comparison, the LHS; otherwise the whole expression. Used if no explicit `the[]`.
        test[the[counter()] < the[counter()]]

    with testset("outer"):
        with testset("inner 1"):
            test[g(6, 7) == 42]
        with testset("inner 2"):
            test[None is None]
        with testset("inner 3"):  # an empty testset is considered 100% passed.
            pass
        with testset("inner 4"):
            warn["This testset not implemented yet"]

    with testset("integration"):
        try:
            import blargly
        except ImportError:
            error["blargly not installed, cannot test integration with it."]
        else:
            ... # blargly integration tests go here

    with testset(postproc=terminate):
        test[2 * 2 == 5]  # fails, terminating the nearest dynamically enclosing `with session`
        test[2 * 2 == 4]  # not reached

We provide the low-level syntactic constructs test[], test_raises[] and test_signals[], with the usual meanings. The last one is for testing code that uses conditions and restarts; see unpythonic.conditions.

The test macros also come in block variants, with test, with test_raises, with test_signals.

As usual in test frameworks, the testing constructs behave somewhat like assert, with the difference that a failure or error will not abort the whole unit (unless explicitly asked to do so).

[docs]

from unpythonic.syntax import macros, let, letseq, letrec

x = let[[a := 1, b := 2] in a + b]
y = letseq[[c := 1,  # LET SEQuential, like Scheme's let*
            c := 2 * c,
            c := 2 * c] in
           c]
z = letrec[[evenp := (lambda x: (x == 0) or oddp(x - 1)),  # LET mutually RECursive, like in Scheme
            oddp := (lambda x: (x != 0) and evenp(x - 1))]
           in evenp(42)]

[docs]

from unpythonic.syntax import macros, dlet

# In Python 3.8, use `@dlet(x << 0)` instead; in Python 3.9, use `@dlet(x := 0)`
@dlet[x := 0]  # let-over-lambda for Python
def count():
    return x := x + 1  # `name := value` rebinds in the let env
assert count() == 1
assert count() == 2

[docs]

from unpythonic.syntax import macros, do, local, delete

x = do[local[a := 21],
       local[b := 2 * a],
       print(b),
       delete[b],  # do[] local variables can be deleted, too
       4 * a]
assert x == 84

[docs]

from unpythonic.syntax import macros, tco

with tco:
    # expressions are automatically analyzed to detect tail position.
    evenp = lambda x: (x == 0) or oddp(x - 1)
    oddp  = lambda x: (x != 0) and evenp(x - 1)
    assert evenp(10000) is True

[docs]

from unpythonic.syntax import macros, autocurry
from unpythonic import foldr, composerc as compose, cons, nil, ll

with autocurry:
    def add3(a, b, c):
        return a + b + c
    assert add3(1)(2)(3) == 6

    mymap = lambda f: foldr(compose(cons, f), nil)
    double = lambda x: 2 * x
    assert mymap(double, (1, 2, 3)) == ll(2, 4, 6)

[docs]

from unpythonic.syntax import macros, lazify

with lazify:
    def my_if(p, a, b):
        if p:
            return a  # b never evaluated in this code path
        else:
            return b  # a never evaluated in this code path
    assert my_if(True, 23, 1/0) == 23
    assert my_if(False, 1/0, 42) == 42

[docs]

from unpythonic.syntax import macros, continuations, call_cc

with continuations:  # enables also TCO automatically
    # McCarthy's amb() operator
    stack = []
    def amb(lst, cc):
        if not lst:
            return fail()
        first, *rest = tuple(lst)
        if rest:
            remaining_part_of_computation = cc
            stack.append(lambda: amb(rest, cc=remaining_part_of_computation))
        return first
    def fail():
        if stack:
            f = stack.pop()
            return f()

    # Pythagorean triples using amb()
    def pt():
        z = call_cc[amb(range(1, 21))]  # capture continuation, auto-populate cc arg
        y = call_cc[amb(range(1, z+1))]
        x = call_cc[amb(range(1, y+1))]
        if x*x + y*y != z*z:
            return fail()
        return x, y, z
    t = pt()
    while t:
        print(t)
        t = fail()  # note pt() has already returned when we call this.

Unpythonic in 30 seconds: Language extensions with dialects

The dialects subsystem of mcpyrate makes Python into a language platform, à la Racket. We provide some example dialects based on unpythonic's macro layer. See documentation.

[docs]

Also comes with automatically named, multi-expression lambdas.

from unpythonic.dialects import dialects, Lispython  # noqa: F401

def factorial(n):
    def f(k, acc):
        if k == 1:
            return acc
        f(k - 1, k * acc)
    f(n, acc=1)
assert factorial(4) == 24
factorial(5000)  # no crash

square = lambda x: x**2
assert square(3) == 9
assert square.__name__ == "square"

# - brackets denote a multiple-expression lambda body
#   (if you want to have one expression that is a literal list,
#    double the brackets: `lambda x: [[5 * x]]`)
# - local[name := value] makes an expression-local variable
g = lambda x: [local[y := 2 * x],
               y + 1]
assert g(10) == 21

[docs]

from unpythonic.dialects import dialects, Pytkell  # noqa: F401

from operator import add, mul

def addfirst2(a, b, c):
    return a + b
assert addfirst2(1)(2)(1 / 0) == 3

assert tuple(scanl(add, 0, (1, 2, 3))) == (0, 1, 3, 6)
assert tuple(scanr(add, 0, (1, 2, 3))) == (0, 3, 5, 6)

my_sum = foldl(add, 0)
my_prod = foldl(mul, 1)
my_map = lambda f: foldr(compose(cons, f), nil)
assert my_sum(range(1, 5)) == 10
assert my_prod(range(1, 5)) == 24
double = lambda x: 2 * x
assert my_map(double, (1, 2, 3)) == ll(2, 4, 6)

[docs]

from unpythonic.dialects import dialects, Listhell  # noqa: F401

from operator import add, mul
from unpythonic import foldl, foldr, cons, nil, ll

(print, "hello from Listhell")

my_sum = (foldl, add, 0)
my_prod = (foldl, mul, 1)
my_map = lambda f: (foldr, (compose, cons, f), nil)
assert (my_sum, (range, 1, 5)) == 10
assert (my_prod, (range, 1, 5)) == 24
double = lambda x: 2 * x
assert (my_map, double, (q, 1, 2, 3)) == (ll, 2, 4, 6)

Install & uninstall

From PyPI

pip install unpythonic

From source

Clone the repo from GitHub. Then, navigate to it in a terminal, and:

pip install . --no-compile

If you intend to use the macro layer of unpythonic, the --no-compile flag is important. It prevents an incorrect precompilation, without macro support, that pip install would otherwise do at its bdist_wheel step.

For most Python projects such precompilation is just fine - it's just macro-enabled projects that shouldn't be precompiled with standard tools.

If --no-compile is NOT used, the precompiled bytecode cache may cause errors such as ImportError: cannot import name 'macros' from 'mcpyrate.quotes', when you try to e.g. from unpythonic.syntax import macros, let. In-tree, it might work, but against an installed copy, it will fail. It has happened that my CI setup did not detect this kind of failure.

This is a common issue when using macro expanders in Python.

Uninstall

pip uninstall unpythonic

Support

Not working as advertised? Missing a feature? Documentation needs improvement?

In case of a problem, see Troubleshooting first. Then:

Issue reports and pull requests are welcome. Contribution guidelines.

While unpythonic is intended as a serious tool for improving productivity as well as for teaching, right now my work priorities mean that it's developed and maintained on whatever time I can spare for it. Thus getting a response may take a while, depending on which project I happen to be working on.

License

All original code is released under the 2-clause BSD license.

For sources and licenses of fragments originally seen on the internet, see AUTHORS.

Acknowledgements

Thanks to TUT for letting me teach RAK-19006 in spring term 2018; early versions of parts of this library were originally developed as teaching examples for that course. Thanks to @AgenttiX for early feedback.

Relevant reading

Links to blog posts, online articles and papers on topics relevant in the context of unpythonic have been collected to a separate document.

If you like both FP and numerics, we have some examples based on various internet sources.