Recently I wrote a function to generate certain sequences with nontrivial constraints. The problem came with a natural recursive solution. Now it happens that, even for relatively small input, the sequences are several thousands, thus I would prefer to use my algorithm as a generator instead of using it to fill a list with all the sequences.

Here is an example. Suppose we want to compute all the permutations of a string with a recursive function. The following naive algorithm takes an extra argument ‘storage’ and appends a permutation to it whenever it finds one:

def getPermutations(string, storage, prefix=""):
   if len(string) == 1:
      storage.append(prefix + string)   # <-----
   else:
      for i in range(len(string)):
         getPermutations(string[:i]+string[i+1:], storage, prefix+string[i])

storage = []
getPermutations("abcd", storage)
for permutation in storage: print permutation

(Please don’t care about inefficiency, this is only an example.)

Now I want to turn my function into a generator, i.e. to yield a permutation instead of appending it to the storage list:

def getPermutations(string, prefix=""):
   if len(string) == 1:
      yield prefix + string             # <-----
   else:
      for i in range(len(string)):
         getPermutations(string[:i]+string[i+1:], prefix+string[i])

for permutation in getPermutations("abcd"):
   print permutation

This code does not work (the function behaves like an empty generator).

Am I missing something? Is there a way to turn the above recursive algorithm into a generator without replacing it with an iterative one?

def getPermutations(string, prefix=""):
    if len(string) == 1:
        yield prefix + string
    else:
        for i in xrange(len(string)):
            for perm in getPermutations(string[:i] + string[i+1:], prefix+string[i]):
                yield perm

Or without an accumulator:

def getPermutations(string):
    if len(string) == 1:
        yield string
    else:
        for i in xrange(len(string)):
            for perm in getPermutations(string[:i] + string[i+1:]):
                yield string[i] + perm

This avoids the len(string)-deep recursion, and is in general a nice way to handle generators-inside-generators:

from types import GeneratorType

def flatten(*stack):
    stack = list(stack)
    while stack:
        try: x = stack[0].next()
        except StopIteration:
            stack.pop(0)
            continue
        if isinstance(x, GeneratorType): stack.insert(0, x)
        else: yield x

def _getPermutations(string, prefix=""):
    if len(string) == 1: yield prefix + string
    else: yield (_getPermutations(string[:i]+string[i+1:], prefix+string[i])
            for i in range(len(string)))

def getPermutations(string): return flatten(_getPermutations(string))

for permutation in getPermutations("abcd"): print permutation

flatten allows us to continue progress in another generator by simply yielding it, instead of iterating through it and yielding each item manually.

Python 3.3 will add yield from to the syntax, which allows for natural delegation to a sub-generator:

def getPermutations(string, prefix=""):
    if len(string) == 1:
        yield prefix + string
    else:
        for i in range(len(string)):
            yield from getPermutations(string[:i]+string[i+1:], prefix+string[i])

The interior call to getPermutations — it’s a generator, too.

def getPermutations(string, prefix=""):
   if len(string) == 1:
      yield prefix + string            
   else:
      for i in range(len(string)):
         getPermutations(string[:i]+string[i+1:], prefix+string[i])  # <-----

You need to iterate through that with a for-loop (see @MizardX posting, which edged me out by seconds!)