Fix cache pollution from mutable reference (getsentry#3887)

- Removes manual overrides of copy behavior and leaves it up to the caller.
    - E.g. a future use case may require a non-deepcopy. If we override copy they would have to remove the dunder copy, update every implementation which relies copy, before finally creating their own copy implementation.
- Deepcopies the flag buffer.
    - Though we do not cache mutable references yet we may soon and so this foot gun should be removed from possibility.
- Removes "copy" test coverage from `test_lru_cache.py`. We're no longer assuming copy usage and leave it up to the caller.
    - The existing test in `tests/test_scope.py` covers the cache pollution case [originally mentioned here](getsentry#3852).
    - The mutable cache pollution case is not covered because we do not currently cache mutable objects.

In general a generic class should assume as few implementation details as possible.  If we leave the existing copy method someone may assume copy semantics and rely on it in a way that is inappropriate.

Closes: getsentry#3886

Co-authored-by: Anton Pirker <[email protected]>

sentry_sdk/_lru_cache.py

@@ -17,15 +17,6 @@ def __init__(self, max_size):
         self.hits = self.misses = 0
         self.full = False
 
-    def __copy__(self):
-        # type: () -> LRUCache
-        new = LRUCache(max_size=self.max_size)
-        new.hits = self.hits
-        new.misses = self.misses
-        new.full = self.full
-        new._data = self._data.copy()
-        return new
-
     def set(self, key, value):
         # type: (Any, Any) -> None
         current = self._data.pop(key, _SENTINEL)

sentry_sdk/flag_utils.py

@@ -1,4 +1,3 @@
-from copy import copy
 from typing import TYPE_CHECKING
 
 import sentry_sdk
@@ -25,12 +24,6 @@ def clear(self):
         # type: () -> None
         self.buffer = LRUCache(self.capacity)
 
-    def __copy__(self):
-        # type: () -> FlagBuffer
-        buffer = FlagBuffer(capacity=self.capacity)
-        buffer.buffer = copy(self.buffer)
-        return buffer
-
     def get(self):
         # type: () -> list[FlagData]
         return [{"flag": key, "result": value} for key, value in self.buffer.get_all()]

sentry_sdk/scope.py

@@ -1,7 +1,7 @@
 import os
 import sys
 import warnings
-from copy import copy
+from copy import copy, deepcopy
 from collections import deque
 from contextlib import contextmanager
 from enum import Enum
@@ -252,7 +252,7 @@ def __copy__(self):
 
         rv._last_event_id = self._last_event_id
 
-        rv._flags = copy(self._flags)
+        rv._flags = deepcopy(self._flags)
 
         return rv
 

tests/test_lru_cache.py

@@ -1,5 +1,4 @@
 import pytest
-from copy import copy, deepcopy
 
 from sentry_sdk._lru_cache import LRUCache
 
@@ -59,55 +58,3 @@ def test_cache_get_all():
     assert cache.get_all() == [(1, 1), (2, 2), (3, 3)]
     cache.get(1)
     assert cache.get_all() == [(2, 2), (3, 3), (1, 1)]
-
-
-def test_cache_copy():
-    cache = LRUCache(3)
-    cache.set(0, 0)
-    cache.set(1, 1)
-
-    copied = copy(cache)
-    cache.set(2, 2)
-    cache.set(3, 3)
-    assert copied.get_all() == [(0, 0), (1, 1)]
-    assert cache.get_all() == [(1, 1), (2, 2), (3, 3)]
-
-    copied = copy(cache)
-    cache.get(1)
-    assert copied.get_all() == [(1, 1), (2, 2), (3, 3)]
-    assert cache.get_all() == [(2, 2), (3, 3), (1, 1)]
-
-
-def test_cache_deepcopy():
-    cache = LRUCache(3)
-    cache.set(0, 0)
-    cache.set(1, 1)
-
-    copied = deepcopy(cache)
-    cache.set(2, 2)
-    cache.set(3, 3)
-    assert copied.get_all() == [(0, 0), (1, 1)]
-    assert cache.get_all() == [(1, 1), (2, 2), (3, 3)]
-
-    copied = deepcopy(cache)
-    cache.get(1)
-    assert copied.get_all() == [(1, 1), (2, 2), (3, 3)]
-    assert cache.get_all() == [(2, 2), (3, 3), (1, 1)]
-
-
-def test_cache_pollution():
-    cache1 = LRUCache(max_size=2)
-    cache1.set(1, True)
-    cache2 = copy(cache1)
-    cache2.set(1, False)
-    assert cache1.get(1) is True
-    assert cache2.get(1) is False
-
-
-def test_cache_pollution_deepcopy():
-    cache1 = LRUCache(max_size=2)
-    cache1.set(1, True)
-    cache2 = deepcopy(cache1)
-    cache2.set(1, False)
-    assert cache1.get(1) is True
-    assert cache2.get(1) is False