Title: hash() and __hash__() do not document their size constraints
Type: behavior Stage:
Components: Documentation Versions: Python 3.4, Python 3.3
Status: closed Resolution: fixed
Dependencies: Superseder:
Assigned To: docs@python Nosy List: barry, docs@python, python-dev
Priority: normal Keywords:

Created on 2013-07-13 18:34 by barry, last changed 2013-07-15 19:23 by barry. This issue is now closed.

Messages (2)
msg193014 - (view) Author: Barry A. Warsaw (barry) * (Python committer) Date: 2013-07-13 18:34
If you have a custom object that implements __hash__() and it returns a value wider than Py_ssize_t, built-in hash() on the object will truncate information.  This is because hash() takes the value returned by obj.__hash__() and coerces it through PyLong_FromSsize_t().  This can cause object hashes to have different values on 64bit and 32bit machines, e.g. on 64bit Linux where Py_ssize_t is 8 bytes wide vs. 32bit Linux where it is 4 bytes wide.

This may be perfectly reasonable from an implementation point of (ref: issue9778) but it is surprising since it is not documented.

This size constraint on object hashes should be documented.

from ctypes import *

class Foo:
    def __hash__(self):
        return 0x1332a6000000000

print(hash(Foo()), sizeof(c_ssize_t))

---64bit Ubuntu 13.10---
$ python3.3
86459409655398400 8

---32bit Ubuntu 13.10---
$ python3
40260800 4
msg193125 - (view) Author: Roundup Robot (python-dev) Date: 2013-07-15 19:22
New changeset 8fe13940b033 by Barry Warsaw in branch '3.3':
- Issue #18440: Clarify that `hash()` can truncate the value returned from an

New changeset f01f0c9cbcc8 by Barry Warsaw in branch 'default':
- Issue #18440: Clarify that `hash()` can truncate the value returned from an
Date User Action Args
2013-07-15 19:23:28barrysetstatus: open -> closed
resolution: fixed
2013-07-15 19:22:36python-devsetnosy: + python-dev
messages: + msg193125
2013-07-15 18:37:18barrysetversions: - Python 3.2
2013-07-13 18:34:45barrycreate