Path.exists() - Memory leak

[RegisGraptin]

Bug report

Bug description:

I am currently having an issue when using Path.exists() function. It seems that if you use it for intensive usage, I have a memory leak.

Here is a python script, where I iterate over some items, and when I called the exists() function it seems the RAM keeps growing.

from collections.abc import Iterator
from pathlib import Path


def check_file(filename: Path) -> None:
 if filename.exists():
 return
 return


def iterate() -> Iterator[int]:
 yield from range(10_000_000)


if __name__ == "__main__":
 directory = Path("/")
 for item in iterate():
 filename = directory / str(item)
 check_file(filename)

Here also a screen shot that I have using memory_profiler library to illustrate it.

I do not know if I am missing something here... Let me know if you need more info to reproduce it. I am using Python 3.12.3 on Ubuntu 22.04.4 LTS.

Thanks

CPython versions tested on:

3.12

Operating systems tested on:

Linux

[ZeroIntensity]

Pathlib interns strings, so that's probably the cause of the inflated memory usage. Duplicate of #119518 (and also take a look at #113993).

[Rogdham]

For what it's worth, the issue seems to be only present in Python 3.12 (not before, nor in 3.13).

Running the following script with docker official slim images:

from pathlib import Path
import tracemalloc

directory = Path("/")
found = False
tracemalloc.start()

snapshot1 = tracemalloc.take_snapshot()
for item in range(1_000_000):
 found |= (directory / str(item)).exists()
snapshot2 = tracemalloc.take_snapshot()

top_stats = snapshot2.compare_to(snapshot1, 'lineno')
for stat in top_stats[:10]:
 print(stat)

• Python 3.9.19

/usr/local/lib/python3.9/pathlib.py:82: size=288 KiB (+288 KiB), count=1 (+1), average=288 KiB
/tmp/docker/m.py:10: size=944 B (+944 B), count=2 (+2), average=472 B
/usr/local/lib/python3.9/tracemalloc.py:423: size=512 B (+512 B), count=3 (+3), average=171 B
/usr/local/lib/python3.9/pathlib.py:976: size=488 B (+488 B), count=1 (+1), average=488 B
/usr/local/lib/python3.9/pathlib.py:738: size=488 B (+488 B), count=1 (+1), average=488 B
/usr/local/lib/python3.9/tracemalloc.py:560: size=472 B (+472 B), count=2 (+2), average=236 B
/usr/local/lib/python3.9/pathlib.py:748: size=464 B (+464 B), count=1 (+1), average=464 B
/usr/local/lib/python3.9/pathlib.py:753: size=456 B (+456 B), count=1 (+1), average=456 B
/usr/local/lib/python3.9/pathlib.py:740: size=456 B (+456 B), count=1 (+1), average=456 B
/usr/local/lib/python3.9/pathlib.py:1426: size=424 B (+424 B), count=1 (+1), average=424 B

• Python 3.10.14

/usr/local/lib/python3.10/pathlib.py:74: size=288 KiB (+288 KiB), count=1 (+1), average=288 KiB
/tmp/docker/m.py:10: size=928 B (+928 B), count=2 (+2), average=464 B
/usr/local/lib/python3.10/tracemalloc.py:423: size=504 B (+504 B), count=3 (+3), average=168 B
/usr/local/lib/python3.10/pathlib.py:855: size=480 B (+480 B), count=1 (+1), average=480 B
/usr/local/lib/python3.10/pathlib.py:617: size=480 B (+480 B), count=1 (+1), average=480 B
/usr/local/lib/python3.10/tracemalloc.py:560: size=464 B (+464 B), count=2 (+2), average=232 B
/usr/local/lib/python3.10/pathlib.py:627: size=456 B (+456 B), count=1 (+1), average=456 B
/usr/local/lib/python3.10/pathlib.py:632: size=448 B (+448 B), count=1 (+1), average=448 B
/usr/local/lib/python3.10/pathlib.py:619: size=440 B (+440 B), count=1 (+1), average=440 B
/usr/local/lib/python3.10/pathlib.py:1290: size=424 B (+424 B), count=1 (+1), average=424 B

• Python 3.11.9

/usr/local/lib/python3.11/pathlib.py:74: size=203 KiB (+203 KiB), count=1 (+1), average=203 KiB
/usr/local/lib/python3.11/tracemalloc.py:560: size=320 B (+320 B), count=2 (+2), average=160 B
/usr/local/lib/python3.11/tracemalloc.py:423: size=320 B (+320 B), count=2 (+2), average=160 B
/tmp/docker/m.py:9: size=32 B (+32 B), count=1 (+1), average=32 B

• Python 3.12.5

<frozen posixpath>:160: size=44.7 MiB (+44.7 MiB), count=999951 (+999951), average=47 B
/usr/local/lib/python3.12/pathlib.py:404: size=29.3 MiB (+29.3 MiB), count=9 (+9), average=3338 KiB
/usr/local/lib/python3.12/tracemalloc.py:560: size=312 B (+312 B), count=2 (+2), average=156 B
/usr/local/lib/python3.12/tracemalloc.py:423: size=312 B (+312 B), count=2 (+2), average=156 B
/tmp/docker/m.py:9: size=32 B (+32 B), count=1 (+1), average=32 B

• Python 3.13.0 rc1

/usr/local/lib/python3.13/pathlib/_local.py:274: size=405 KiB (+405 KiB), count=1 (+1), average=405 KiB
/usr/local/lib/python3.13/tracemalloc.py:560: size=328 B (+328 B), count=1 (+1), average=328 B
/usr/local/lib/python3.13/tracemalloc.py:423: size=328 B (+328 B), count=1 (+1), average=328 B
/tmp/docker/m.py:9: size=32 B (+32 B), count=1 (+1), average=32 B

[Rogdham]

Possibly closer duplicate of #121780, with patch in #120520 that has been backported to 3.13 but not 3.12 it seems.

[ZeroIntensity]

@encukou, was #120520 supposed to be backported to 3.12?

[encukou]

Yes, but it's a big PR and there are several fixes on top. I'm now working on the backport in #123065.

[rafsaf]

Possibly closer duplicate of #121780, with patch in #120520 that has been backported to 3.13 but not 3.12 it seems.

Small note: while patched to "standard" 3.13, on experimental free-threaded build 3.13.0rc2t that is still the case - memory usage increases indefinitely because of interned strings use in pathlib _parse_path method. But afaik and from what I understand, at least there it is expected for now and probably will be addressed in pathlib code itself in the future or handled somehow :p

[rafsaf]

@RegisGraptin This was fixed in 3.12.7 release on October 1st.

https://docs.python.org/release/3.12.7/whatsnew/changelog.html#python-3-12-7

The issue can be closed. Thanks!

[RegisGraptin]

Thank you for the work 🙏 , doubled tested on my side on 3.12.7 with the code of the issue, and no more memory leak noticed 🎉