Python: What is the hard recursion limit for Linux

问题:

Python's sys module provides a function setrecursionlimit that lets you change Python's maximum recursion limit. The docs say:

The highest possible limit is platform-dependent.

My question is: What is the highest possible limits for various platforms, under CPython? I would like to know the values for Linux, Mac and Windows.

UPDATE: Can we please avoid "You're doing it wrong" answers? I know that trying to do very deep recursion is usually a bad idea. I've considered the pros and cons in my specific situation and decided that I want to do it.

回答1:

On Windows (at least), sys.setrecursionlimit isn't the full story. The hard limit is on a per-thread basis and you need to call threading.stack_size and create a new thread once you reach a certain limit. (I think 1MB, but not sure) I've used this approach to increase it to a 64MB stack.

import sys
import threading

threading.stack_size(67108864) # 64MB stack
sys.setrecursionlimit(2 ** 20) # something real big
                               # you actually hit the 64MB limit first
                               # going by other answers, could just use 2**32-1

# only new threads get the redefined stack size
thread = threading.Thread(target=main)
thread.start()

I haven't tried to see what limits there might be on threading.stack_size, but feel free to try... that's where you need to look.

In summary, sys.setrecursionlimit is just a limit enforced by the interpreter itself. threading.stack_size lets you manipulate the actual limit imposed by the OS. If you hit the latter limit first, Python will just crash completely.

回答2:

Default values for major operating systems;

• For Windows: 2000
• For Linux: 1000
• For Mac OS: 1000

回答3:

You shouldn't overuse recursive calls in CPython. It has not tail optimization, the function calls use a lot of memory and processing time. Those limits might not apply to other implementations, it's not in the blueprints.

In CPython, recursion is fine for traversing data structures (where a limit of 1000 should be enough for everybody) but not for algorithms. If I were to implement, say, graph related algorithms and hit the recursion limit, I would either implement my own stack and use iterations, or look for libraries implemented in C/C++/whatever before raising the limit by hand.