I just finished the lecture on streams in SICP. Python has generators, which are quite lovely. However, it's fun to imagine reimplementing generators in various ways from scratch using some of the approaches in SICP.
First of all, there are promises. A promise says that you'll do something later. In Scheme, you create a promise with "delay" and you invoke that promise with "force":
Now, let's take a look at a generator. Here's a simple counter:
First of all, there are promises. A promise says that you'll do something later. In Scheme, you create a promise with "delay" and you invoke that promise with "force":
> (delay (/ 1 0))Here's one way to translate that to Python:
#<promise>
> (force (delay (/ 1 0)))
division by zero
class Promise(object):Easy :)
"""This is a promise to execute something on demand."""
def __init__(self, f):
"""Wrap your expression in a lambda and pass it as f."""
self.f = f
def __call__(self):
"""Call the instance to get the value of the promise.
This is memoized so that it doesn't have to compute the value
more than once since promises shouldn't do that.
"""
if not hasattr(self, 'result'):
self.result = self.f()
return self.result
promise = Promise(lambda: 1 / 0)
print repr(promise)
print "No error yet."
print "Now, we'll force the promise, causing the error."
promise()
Now, let's take a look at a generator. Here's a simple counter:
def generator(start=0):Clearly, you can do this with an iterator class. However, let's instead use a closure and continuation passing style. The syntax isn't the most beautiful in the world, but it definitely does the trick ;)
while True:
yield start
start += 1
for i in generator():
print i
def generator(start=0):
scope = locals()
def top():
while True:
return scope['start'], bottom
def bottom():
scope['start'] += 1
return top()
return top
next = generator()
while True:
(i, next) = next()
print i
Comments
Not sure whether you've seen this before, but here's a similar post from Tim Peters on the Python mailing list from 2000: http://mail.python.org/pipermail/python-list/2000-February/025636.html
It obviously predates generators in Python, as well as nested lexical scopes (as you can see in the i=i, inc=inc argument passing). But it's interesting how the basic semantics are very similar to your example.
(Afaik, Python's generators were borrowed from Icon, but I don't have time to go dig up references today.)
(e.g.
a = promise()
a.foo = "bar"
b = promise()
# b == a, not the original value of the
# expression
)
On the other hand, caching might be well motivated because the expression might have a side effect, which we only expect to trigger once when we set up the promise.
So Lisp inspired Python is a bit less "functional" than the original Lisp?
That's awesome! We probably got it from the same book. For some reason, he's not memoizing inside the promise. Oh well.
I really like his Fibonacci example:
fibs = Stream(1,
lambda: Stream(1,
lambda: sadd(fibs,
fibs.tail())))
He mentions Haskell, so he was probably referring to the classic Haskell example:
fib = 1 : 1 : [ a+b | (a,b) <- zip fib (tail fib) ]
Of course not! I'm implying that since you can build everything in assembly, everything that has something is automatically inspired by assembly!
It just amazes me to see this stuff in a video from 1986 ;)
That's actually the desired result.
I agree with you, however, side-effects make everything more complex.