Yield in a recursive function

Try this:

if stat.S_ISDIR(stat_info.st_mode):
    for p in explore(path):
        yield p

That calls explore like a function. What you should do is iterate it like a generator:

if stat.S_ISDIR(stat_info.st_mode):
  for p in explore(path):
    yield p
else:
  yield path

EDIT: Instead of the stat module, you could use os.path.isdir(path).

Iterators do not work recursively like that. You have to re-yield each result, by replacing

explore(path)

with something like

for value in explore(path):
    yield value

Python 3.3 added the syntax yield from X, as proposed in PEP 380, to serve this purpose. With it you can do this instead:

yield from explore(path)

If you’re using generators as coroutines, this syntax also supports the use of generator.send() to pass values back into the recursively-invoked generators. The simple for loop above would not.

Change this:

explore(path)

To this:

for subpath in explore(path):
    yield subpath

Or use os.walk, as phooji suggested (which is the better option).

The problem is this line of code:

explore(path)

What does it do?

• calls explore with the new path
• explore runs, creating a generator
• the generator is returned to the spot where explore(path) was executed . . .
• and is discarded

Why is it discarded? It wasn’t assigned to anything, it wasn’t iterated over — it was completely ignored.

If you want to do something with the results, well, you have to do something with them! 😉

The easiest way to fix your code is:

for name in explore(path):
    yield name

When you are confident you understand what’s going on, you’ll probably want to use os.walk() instead.

Once you have migrated to Python 3.3 (assuming all works out as planned) you will be able to use the new yield from syntax and the easiest way to fix your code at that point will be:

yield from explore(path)

os.walk is great if you need to traverse all the folders and subfolders. If you don’t need that, it’s like using an elephant gun to kill a fly.

However, for this specific case, os.walk could be a better approach.

You can also implement the recursion using a stack.

There is not really any advantage in doing this though, other than the fact that it is possible. If you are using python in the first place, the performance gains are probably not worthwhile.

import os
import stat

def explore(p):
    '''
    perform a depth first search and yield the path elements in dfs order
        -implement the recursion using a stack because a python can't yield within a nested function call
    '''
    list_t=type(list())
    st=[[p,0]]
    while len(st)>0:
        x=st[-1][0]
        print x
        i=st[-1][1]

        if type(x)==list_t:
            if i>=len(x):
                st.pop(-1)
            else:
                st[-1][1]+=1
                st.append([x[i],0])
        else:
            st.pop(-1)
            stat_info = os.lstat(x)
            if stat.S_ISDIR(stat_info.st_mode):
                st.append([['%s/%s'%(x,a) for a in os.listdir(x)],0])
            else:
                yield x

print list(explore('.'))

To answer the original question as asked, the key is that the yield statement needs to be propagated back out of the recursion (just like, say, return). Here is a working reimplementation of os.walk(). I’m using this in a pseudo-VFS implementation, where I additionally replace os.listdir() and similar calls.

import os, os.path
def walk (top, topdown=False):
    items = ([], [])
    for name in os.listdir(top):
        isdir = os.path.isdir(os.path.join(top, name))
        items[isdir].append(name)
    result = (top, items[True], items[False])
    if topdown:
        yield result
    for folder in items[True]:
        for item in walk(os.path.join(top, folder), topdown=topdown):
            yield item
    if not topdown:
        yield result