functools.cache vs sys.setrecursionlimit

[gsbrodal]

Bug report

Bug description:

When running the below code with Python 3.13.0 on Windows

import functools
import sys

sys.setrecursionlimit(1_000_000)

@functools.cache
def double(n):
    return 2 + double(n - 1) if n > 0 else 0

print(f'{double(2_000) = }')

I get the following error

return 2 + double(n - 1) if n > 0 else 0
           ~~~~~~^^^^^^^
[Previous line repeated 996 more times]
RecursionError: maximum recursion depth exceeded"

It appears that the sys.setrecursionlimit does not carry over to functools.cache. The error does not occur when using Python 3.12.3.

CPython versions tested on:

3.13

Operating systems tested on:

Windows

[Eclips4]

Duplicate of #112215

[gsbrodal]

Sorry for the slightly wrong info - the problem also persisted in Python 3.12 on Windows (I was not running the program with Python 3.12 on my Windows machine). I just downloaded the Windows binaries for Python 3.10.10, 3.11.7, and 3.12.7, and all have problems with increasing the recursionlimit and the functools.cache.

I know increasing the recursion limit comes with a risk, so I ask for trouble. I was just hopping that I always could use functools.cache instead of writing my own @cache decorator, like the below. The below @cache decorator handles significantly higher recursion depths with Python 3.13.0 on Windows (and with 3.12 and 3.11, but not with 3.10).

import sys
sys.setrecursionlimit(1_000_000)

def cache(f):
    answers = {}
    def wrapper(x):
        if x not in answers:
            answers[x] = f(x)
        return answers[x]
    return wrapper

@cache
def double(n):
    return 2 + double(n - 1) if n > 0 else 0

print(f'{double(400_000) = }')  # prints 800_000

[Eclips4]

Sorry for the slightly wrong info - the problem also persisted in Python 3.12 on Windows (I was not running the program with Python 3.12 on my Windows machine). I just downloaded the Windows binaries for Python 3.10.10, 3.11.7, and 3.12.7, and all have problems with increasing the recursionlimit and the functools.cache.

I know increasing the recursion limit comes with a risk, so I ask for trouble. I was just hopping that I always could use functools.cache instead of writing my own @cache decorator, like the below. The below @cache decorator handles significantly higher recursion depths with Python 3.13.0 on Windows (and with 3.12 and 3.11, but not with 3.10).

import sys
sys.setrecursionlimit(1_000_000)

def cache(f):
    answers = {}
    def wrapper(x):
        if x not in answers:
            answers[x] = f(x)
        return answers[x]
    return wrapper

@cache
def double(n):
    return 2 + double(n - 1) if n > 0 else 0

print(f'{double(400_000) = }')  # prints 800_000

Yeah, I understand your concerns, Gerth. Can you please continue discussion on #112215? Thanks in advance!