Pure Python pickle.py Raises Wrong Exceptions for Invalid Data
Nadie ha tomado este issue todavía.
- Lenguaje dominante
- Python
- Estrellas
- 77.2k
- Forks
- 35.9k
- Métricas de merge de PR
- Métricas de PR pendientes
Descripción
Bug report
Bug description:
Summary
The pure Python implementation of pickle.py in CPython's standard library raises incorrect exceptions (KeyError and IndexError) when unpickling invalid data, instead of the expected pickle.UnpicklingError. This bug is masked in normal CPython usage because the C implementation (_pickle) is used by default, but it affects any Python implementation that relies on the pure Python fallback (e.g., Brython, PyPy in certain modes).
Issue Details
When the pure Python _Unpickler class encounters invalid pickle data, it fails to provide proper error handling in two scenarios:
Problem 1: Invalid Opcodes (KeyError)
When an unrecognized opcode is encountered, dispatch[key[0]](self) raises KeyError instead of UnpicklingError.
Expected behavior: pickle.UnpicklingError: invalid load key, '\xff'.
Actual behavior: KeyError: 255
Problem 2: Missing MARK Instructions (IndexError)
When opcodes like TUPLE are encountered without a preceding MARK instruction, pop_mark() attempts to pop from an empty metastack, raising IndexError instead of UnpicklingError.
Expected behavior: pickle.UnpicklingError: could not find MARK
Actual behavior: IndexError: pop from empty list
Steps to Reproduce
Reproducing in CPython (Force Pure Python Implementation)
The bug exists in CPython's pickle.py but is not normally visible because CPython uses the C _pickle module. To reproduce, you must disable the C implementation:
# test_pickle_bug.py
import sys
# Block the C implementation to force pure Python fallback
sys.modules['_pickle'] = None
import pickle
print("Test 1: Invalid opcode (should raise UnpicklingError)")
try:
pickle.loads(b'\xff\xff\xff')
print("FAIL: Expected UnpicklingError")
except pickle.UnpicklingError as e:
print(f"PASS: {e}")
except KeyError as e:
print(f"FAIL: Got KeyError instead: {e}")
print("\nTest 2: Missing MARK (should raise UnpicklingError)")
try:
pickle.loads(b'this is not valid pickle data')
print("FAIL: Expected UnpicklingError")
except pickle.UnpicklingError as e:
print(f"PASS: {e}")
except IndexError as e:
print(f"FAIL: Got IndexError instead: {e}")
Expected output (with C _pickle):
Test 1: Invalid opcode (should raise UnpicklingError)
PASS: invalid load key, '\xff'.
Test 2: Missing MARK (should raise UnpicklingError)
PASS: could not find MARK
Actual output (with pure Python pickle.py):
Test 1: Invalid opcode (should raise UnpicklingError)
FAIL: Got KeyError instead: 255
Test 2: Missing MARK (should raise UnpicklingError)
FAIL: Got IndexError instead: pop from empty list
Root Cause Analysis
Location 1: _Unpickler.load() method
File: Lib/pickle.py
Approximate line: 1212 (may vary by Python version)
try:
while True:
key = read(1)
if not key:
raise EOFError
assert isinstance(key, bytes_types)
dispatch[key[0]](self) # <-- KeyError raised here for invalid opcodes
except _Stop as stopinst:
return stopinst.value
The dispatch dictionary lookup is not wrapped in error handling, so KeyError propagates directly to the caller instead of being converted to UnpicklingError.
Location 2: _Unpickler.pop_mark() method
File: Lib/pickle.py
Approximate line: 1219 (may vary by Python version)
def pop_mark(self):
items = self.stack
self.stack = self.metastack.pop() # <-- IndexError raised here if metastack is empty
self.append = self.stack.append
return items
When pop_mark() is called without a preceding MARK instruction, the metastack is empty, causing list.pop() to raise IndexError.
Proposed Fix
Add proper error handling to both locations:
Fix 1: Handle KeyError in load() method
try:
while True:
key = read(1)
if not key:
raise EOFError
assert isinstance(key, bytes_types)
try:
dispatch[key[0]](self)
except KeyError:
raise UnpicklingError(
f"invalid load key, '\\x{key[0]:02x}'.")
except _Stop as stopinst:
return stopinst.value
Fix 2: Handle empty metastack in pop_mark() method
def pop_mark(self):
if not self.metastack:
raise UnpicklingError("could not find MARK")
items = self.stack
self.stack = self.metastack.pop()
self.append = self.stack.append
return items
Impact
Who is affected?
- Alternative Python implementations that cannot use the C
_picklemodule (e.g., Brython, potentially PyPy in pure mode, MicroPython if it includes pickle) - Security tools that analyze pickle data and expect proper exception types
- Testing frameworks that validate pickle error handling
- Code that explicitly uses pure Python pickle for debugging or compatibility
Why this matters?
- API consistency: The pure Python implementation should behave identically to the C implementation
- Exception handling: Code that catches
pickle.UnpicklingErrorfor malformed data will fail to catch these errors - Security: Malformed pickle data should be handled gracefully with appropriate error messages
- Debugging: Proper error messages help developers identify pickle format issues
Verification
After applying the fix, both test cases should pass:
import sys
sys.modules['_pickle'] = None
import pickle
# Test 1: Invalid opcode
try:
pickle.loads(b'\xff\xff\xff')
assert False, "Should raise UnpicklingError"
except pickle.UnpicklingError as e:
assert "invalid load key" in str(e)
print("✓ Test 1 passed")
# Test 2: Missing MARK
try:
pickle.loads(b'this is not valid pickle data')
assert False, "Should raise UnpicklingError"
except pickle.UnpicklingError as e:
assert "could not find MARK" in str(e)
print("✓ Test 2 passed")
# Test 3: Using mock file (from unittest.mock)
from io import BytesIO
from unittest.mock import patch, mock_open
invalid_data = b"invalid pickle"
with patch("builtins.open", mock_open(read_data=invalid_data)):
with open("dummy", "rb") as f:
try:
pickle.load(f)
assert False, "Should raise UnpicklingError"
except pickle.UnpicklingError:
print("✓ Test 3 passed")
Additional Context
This issue was discovered while fixing pickle compatibility in Brython. The bug probably exists in all versions of CPython but has gone unnoticed because:
- CPython always uses the C
_picklemodule in practice - The pure Python
pickle.pyis rarely executed - Most developers don't test with malformed pickle data
However, the pure Python implementation is part of the standard library API contract and should behave identically to the C implementation.
Related Code Patterns
Note that CPython's pickle.py already has precedent for this type of error handling. For example, the load_get(), load_binget(), and load_long_binget() methods all wrap KeyError exceptions when accessing the memo dictionary:
def load_binget(self):
i = self.read(1)[0]
try:
self.append(self.memo[i])
except KeyError as exc:
msg = f'Memo value not found at index {i}'
raise UnpicklingError(msg) from None
The proposed fixes follow this same established pattern.
Python Versions Affected
All Python versions with pure Python pickle.py implementation (verified on 3.9 and 3.14, likely affects 3.9+).
CPython versions tested on:
3.14
Operating systems tested on:
macOS
Linked PRs
- gh-141751
- gh-141754
Guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Línea de trabajo
Comienza con Lib/pickle.py, centrándote en _Unpickler.load() y pop_mark(), y reproduce los dos casos de entrada malformada con el fallback de Python puro descrito en el issue. Se considera terminado cuando los opcodes no válidos y las instrucciones MARK ausentes generen de forma coherente pickle.UnpicklingError con los mensajes esperados; el issue incluye las PRs enlazadas gh-141751 y gh-141754, por lo que es posible que el trabajo ya esté en curso.
Escrito por el modelo de indexación a partir del texto del issue.
Evaluación
- Stack tecnológico
- python
- Área
- backend
- Tipo de issue
- Error
- Dificultad
- 3/5
- Tiempo estimado
- 1-2 días
- Estado de actividad
- Estancado
- Claridad
- Bien especificado
- Aptitud para principiantes
- 25/100