### 技术控

今日:68| 主题:51223

# [其他] The stack of iterators pattern.

163 3

### 立即注册CoLaBug.com会员，免费获得投稿人的专业资料，享用更多功能，玩转个人品牌！

x
The stack of iterators pattern

Gareth Rees , 2016-09-28
Depth-first search  is a straightforward algorithm for visiting the nodes of a tree or tree-like data structure.

Here’s how you might implement it in Python:
[code]def search(node):
for child in children(node):
search(child)[/code]  This works well in many cases, but it has a few problems:

•   It descends the tree using a function call search(child) , and function calls are quite slow in Python.

•   It descends the tree by recursing, and so it uses as many levels of stack as the depth of the deepest node in the tree. But Python’s call stack is limited in size (see  sys.getrecursionlimit  ) and so deep enough trees will run out of call stack and fail with “RuntimeError: maximum recursion depth exceeded.”

• If you’d like to be able to stop the search part way through and return a result (for example returning a target node as soon as you find it), then this requires a slightly awkward change to the code. You could add logic for exiting the recursion:
[code]def search(node):
if target(node):
return node
for child in children(node):
result = search(child)
if result is not None:
return result[/code] or you could return the result non-locally using an exception:
[code]class Found(Exception):
pass

def search(node):
if target(node):
raise Found(node)
for child in children(node):
search(child)[/code] or you could rewrite the search to use generators:
[code]def search(node):
if target(node):
yield node
for child in children(node):
yield from search(child)[/code]  and then the caller can call next(search(root)) .

These problems can all be avoided using the stack of iterators design pattern.
[code]def search(root):
stack = [iter([root])]
while stack:
for node in stack[-1]:
stack.append(iter(children(node)))
break
else:
stack.pop()[/code]  This avoids the three problems above:

• Pushing and popping a list is faster than calling a function in Python.

• Lists can grow without limit, unlike the function call stack.

• Since there’s no recursion, you can just return when you have a result:
[code]def search(root):
stack = [iter([root])]
while stack:
for node in stack[-1]:
if target(node):
return node
stack.append(iter(children(node)))
break
else:
stack.pop()[/code]
The control flow here might seem a bit tricky if you’re not used to the way that Python’s for ... else construct interacts with break . The pattern works by maintaining a stack of iterators that remember the position reached in the iteration over the children of the node at each level of the search. After pushing a new iterator on the stack, the break exits the for loop, bypasses the else: clause, and goes round the while loop again, so that it picks up the new iterator from stack[-1] . When there are no more children in the current iteration, the for loop exits via the else: clause and pops the stack. Then the next iteration of the while loop picks up the iteration at the previous level from where it left off.
✴
Two examples. First, finding a key in a possibly nested dictionary:
[code]def search(d, key, default=None):
"""Return a value corresponding to the specified key in the (possibly
nested) dictionary d. If there is no item with that key, return
default.

"""
stack = [iter(d.items())]
while stack:
for k, v in stack[-1]:
if isinstance(v, dict):
stack.append(iter(v.items()))
break
elif k == key:
return v
else:
stack.pop()
return default[/code]   Second, finding a simple path visiting a set of positions on a grid:

[code]def hamilton_path(start, positions, directions=((0, 1), (1, 0), (0, -1), (-1, 0))):
"""Find a simple path that visits all positions.

start: tuple(int, int)
Starting position for the path.
positions: iterable of tuple(int, int)
Iterable of positions to be visited by the path.
directions: iterable of tuple(int, int)
Iterable of directions to take at each step.

Return the path as a list of tuple(int, int) giving the order in
which the positions are visited. Raise ValueError if there are no
paths visiting all positions.

"""
positions = set(positions)
directions = list(directions)
path = [start]
stack = [iter(directions)]
while path:
x, y = path[-1]
for dx, dy in stack[-1]:
pos = x + dx, y + dy
if pos in positions:
path.append(pos)
positions.remove(pos)
stack.append(iter(directions))
if not positions:
return path
break
else:
positions.add(path.pop())
stack.pop()
raise ValueError("no path")[/code]

•   This requires Python 3.3 or later in order to be able to use the  yield from  statement.

•   In software, a  design pattern  is a technique for working around a defect or omission in a particular programming language. For example, the well-known singleton pattern is a work-around for the lack of global variables in the Java language.

•   Example adapted from this answer on Code Review.

•   Example adapted from this answer on Code Review.

 矿难在检讨中继续，楼价在控制中上升。

 爱情就像两个拉着橡皮筋的人,受伤的总是不愿意放手的那一个!

 楼主，笑一个，萌萌哒！

*滑动验证: #conqu3r_oc_viewthread_fastpost_content .clickCaptcha{top:-270px;}

• ## 2017支付宝集五福在哪扫福字怎么扫福字 支

支付宝在哪扫福字?2017集福怎么扫福字?今天支付 [...]

• ## 2017支付宝福卡怎么获得 2017支付宝如何集

2017支付宝福卡怎么获得呢?本文小编为您带来2017 [...]

• ## 在百度，有哪件大事是非陆奇不可的？

昨天，雷锋网 (公众号：雷锋网) 第一件事 [...]

• ## 亲测有效之后 邮政将正式接手无人机邮路的

杭垓镇位于浙江省安吉县西部，全镇辖区面积266.5 [...]

• ## 口袋记账CEO：2016年我们是如何赢得市场地

本文为口袋记账CEO徐江涛受邀2017第二届移动互 [...]

• ## Practical JSONP Injection

JSONP injection is a lesser known but quite w [...]

• ## 网商贷3天无理由退息怎么申请？

网商贷3天无理由退息是什么?“3天无理由退息”是 [...]

• ## 怎么找回删除的微信聊天记录：苹果安卓手机

恢复删除的微信聊天记录是智能手机兴起后新兴的 [...]

• ## 斯坦福大学开启项目：寻找Apple Watch在健

斯坦福大学最近宣布了一个新的研究项目，目的是 [...]

• ## 酷派：六名员工并未将资料带出华为并交给酷

今日，有媒体报道称，华为内部通报，六名前中高 [...]

© 2001-2017 Comsenz Inc. Design: Dean. DiscuzFans.