Memory Leak on MacOS

[peterallenwebb]

Calling the keyring.get_password() function repeatedly on MacOS results in leaked memory. The problem appears to be in find_generic_password(), where the memory allocated by SecItemCopyMatching() is never released. I spent some time trying to fix the issue, but it was dragging me further into the world of CoreFoundation internals than I had time to go.

I don't know how serious the problem is in practice, but using memray I was seeing a non-negligible memory allocation (~30MB) hanging around in my application, attributable to this function.

Thank you so much for your work maintaining this library! If I have more time soon I might return to the issue and attempt a fix.

[jaraco]

Thanks for reporting. Sounds like a bug to me. I'm also not exactly familiar with the object lifecycle management in CoreFoundation, so I welcome your contrib (or someone else). If someone starts on it, let me know so I can assign the issue to you and avoid duplication of work.

[patrick-kidger]

Hi folks. I've just bumped into this, and I think I've tracked this down. MWE:

import gc
import weakref

ref = None

class MyVeryLargeObject:
    pass

def f():
    global ref
    x = MyVeryLargeObject()
    ref = weakref.ref(x)
    import keyring.backends.Windows

f()
gc.collect()
print(ref())  # <__main__.MyVeryLargeObject object at 0x101611f10>

The issue is because this exception trap:

keyring/keyring/backends/Windows.py

Line 12 in edaa964

with ExceptionTrap() as missing_deps:

is holding on to the exception at the global level forever. This exception keeps a reference to the current traceback, and that tracebacks keeps references to all the local variables up the stack.

Replacing import keyring.backends.Windows with keyring.get_password() reproduces the same behaviour, I assume it's doing a dynamic import internally.

I'm not suitably familiar with keyring to offer a PR, but hopefully this helps :)