Pickling / Unpickling leads to inconsistent hash value

[matthiasdiener]

Running the following program twice can lead to inconsistent hash values due to restoring the hash from the pickled file (which might differ from the hash value when creating the same immutabledict from scratch):

from pickle import dump, load
from immutabledict import immutabledict

i1  = immutabledict({"a": "1", "b": "2", "c": "3"})
hash(i1)  # Force creating a cached hash value

try:
    # Second run: load pickle file
    with open("pickle_imm.pkl", "rb") as f:
        i2 = load(f)
except FileNotFoundError:
    # First run: create pickle file
    i2 = immutabledict({"a": "1", "b": "2", "c": "3"})
    with open("pickle_imm.pkl", "wb") as f:
        dump(i1, f)

assert i1 == i2
print(hash(i1), hash(i2))
assert hash(i1) == hash(i2)  # Fails on the second run: immutabledicts compare equal, but their hash values are different!

$ python pickletest.py
1128449412753896495 1128449412753896495
$ python pickletest.py
-5549929714807656641 1128449412753896495
Traceback (most recent call last):
  File "/Users/mdiener/Work/orderedsets/pickletest.py", line 24, in <module>
    assert hash(i1) == hash(i2)
           ^^^^^^^^^^^^^^^^^^^^
AssertionError

This mostly affects situations where strings, class types, etc. are stored in the immutabledict, which do not have stable hash values across Python invocations.

There are a few potential solutions:

• Exclude the cached hash from __getstate__/__setstate__
• Use pytools.memoize_method to cache the hash value (similar to what I did in use pytools.memoize for caching matthiasdiener/orderedsets#28)
• Use something like https://stackoverflow.com/a/71663059/1250282 to cache the hash value
• Don't cache the hash value at all