gh-141749: align exceptions raised by pickle.load with _pickle.load

[djoume]

Disclaimer: I used Claude Code to help me make this change

This fix addresses error handling inconsistencies in CPython's pure Python pickle.py implementation to match the behavior of the C _pickle module. The changes make the pure Python implementation raise proper UnpicklingError exceptions for invalid pickle data instead of low-level KeyError and IndexError exceptions.

• Invalid opcodes now raise UnpicklingError instead of KeyError
• Operations requiring MARK (TUPLE, LIST, DICT, etc.) now raise UnpicklingError: could not find MARK instead of IndexError: pop from empty list

Note: the C implementation (_pickle) always raises UnpicklingError for all error conditions because it has explicit error checking. The pure Python implementation can't catch all cases without significant performance overhead, so there are still some cases where the pure Python implementation will raise IndexError (e.g., self.stack.pop() on empty stack, self.stack[-1] on empty stack)

This approach:

• ✅ Improves error messages for common cases
• ✅ Maintains performance (no try-except around every stack operation)
• ✅ Passes all existing tests
• ✅ Makes pure Python implementation more consistent with C implementation
• ✅ Follows the precedent set in the code (e.g., load_get, load_binget catch KeyError for memo access)

What Gets Fixed

Before the fix:

>>> pickle.loads(b'\xff\xff\xff')
KeyError: 255

>>> pickle.loads(b'this is not valid pickle data')
IndexError: pop from empty list

After the fix:

>>> pickle.loads(b'\xff\xff\xff')
pickle.UnpicklingError: invalid load key, '\xff'.

>>> pickle.loads(b'this is not valid pickle data')
pickle.UnpicklingError: could not find MARK

Compatibility

• ✅ Backward compatible: Code catching UnpicklingError continues to work
• ✅ Forward compatible: Code catching IndexError continues to work (for other stack operations)
• ✅ All existing CPython tests pass
• ✅ Behavior is closer to C implementation

Impact

This fix benefits:

• Brython: Primary use case - needs pure Python pickle
• PyPy: When using pure Python mode
• MicroPython: If it includes pickle
• Jython: If using CPython stdlib
• Debugging: Better error messages for developers
• Security tools: Proper exception types for malformed pickle validation

• Issue: Pure Python pickle.py Raises Wrong Exceptions for Invalid Data #141749

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[picnixz]

Do not catch the KeyError for the call. Store the dispatcher separately.

[picnixz]

Do we still need to catch IndexError?

[djoume]

Yes, IndexError is still needed. This PR only fixes specific cases (invalid opcodes, missing MARK). Many other operations like self.stack[-1], self.stack.pop() can still raise IndexError on an empty stack during normal unpickling operations.

[picnixz]

Suggested change

	Pure python pickle.py error handling is more consistent with the c implementation, raising UnpicklingError exceptions for invalid pickle data instead of KeyError or IndexError.
	:mod:`pickle`: align exceptions raised by the pure Python implementation of :func:`pickle.load`
	with the C implementation. Previous cases raising :exc:`KeyError` or :exc:`IndexError`
	now raise :exc:`~pickle.UnpicklingError`.

You can also add "Patch by [your name]" at the end.