Test failure in test_hash_equal on py2.6 windows

[agroszer]

Commit 2a8313c breaks tests on winbot

see here:

http://winbot.zope.org/builders/persistent_py_265_32/builds/971/steps/test/logs/stdio

[agroszer]

@jamadden ^

[mgedmin]

I have a feeling this is because the tests run on a 32-bit build rather than because it's Windows or because it's Python 2.6.

Note that hash('a string') will differ on 32- and 64-bit platforms (and also different Python versions may compute different hashes):

>>> hash('a string')  # Python 2.6.5, Linux, 32-bit
1827144452
>>> hash('a string')  # Python 2.6.9, Linux, 64-bit
-8411828025894108412
>>> hash('a string')  # Python 3.4.2, Linux, 64-bit
-8227545103199816438

[mgedmin]

At some point Travis CI might allow us to test both 32- and 64-bit builds in the build matrix: travis-ci/travis-ci#986

[mgedmin]

Okay, string hashes are irrelevant for this issue:

persistent/persistent/_timestamp.c

Lines 155 to 167 in 663497d

	static long
	TimeStamp_hash(TimeStamp *self)
	{
	register unsigned char *p = (unsigned char *)self->data;
	register int len = 8;
	register long x = *p << 7;
	while (--len >= 0)
	x = (1000003*x) ^ *p++;
	x ^= 8;
	if (x == -1)
	x = -2;
	return x;
	}

[mgedmin]

Perhaps it would be sufficient to replace ctypes.c_int64 with ctypes.c_long in pyTimeStamp.hash?

[mgedmin]

Unfortunately the test still fails: http://winbot.zope.org/builders/persistent_py_265_32/builds/972/steps/test/logs/stdio

Failure in test test_hash_equal_constants (persistent.tests.test_timestamp.PyAndCComparisonTests)
Traceback (most recent call last):
  File "c:\Python26_32\lib\unittest.py", line 279, in run
    testMethod()
  File "c:\buildslave\persistent\build\persistent\tests\test_timestamp.py", line 302, in test_hash_equal_constants
    self.assertEqual(hash(c), 1000006000001)
  File "c:\Python26_32\lib\unittest.py", line 350, in failUnlessEqual
    (msg or '%r != %r' % (first, second))
AssertionError: -721379967 != 1000006000001L

Failure in test test_py_hash_32_64_bit (persistent.tests.test_timestamp.PyAndCComparisonTests)
Traceback (most recent call last):
  File "c:\Python26_32\lib\unittest.py", line 279, in run
    testMethod()
  File "c:\buildslave\persistent\build\persistent\tests\test_timestamp.py", line 275, in test_py_hash_32_64_bit
    self.assertEqual(hash(py), bit_64_hash)
  File "c:\Python26_32\lib\unittest.py", line 350, in failUnlessEqual
    (msg or '%r != %r' % (first, second))
AssertionError: 1979033151 != -3850693964765720575L

[mgedmin]

The 1st failure (test_hash_equal_constants) is the same as before. The failure in test_hash_equal is gone. The failure in test_py_hash_32_64_bit is new.

[jamadden]

In test_py_hash_32_64_bit, I probably need to directly call __hash__ to avoid the interpreter's built-in truncating of the 64-bit value.

As for the constants, sorry I missed those the first time around. I obviously calculated those based on a 64-bit platform. What if we skip that test on a 32-bit platform?

[jamadden]

By the way, these failures don't go to the "zope-dev" mailing list, like, for example, the zope.toolkit failures do. Is there some other list they go to so I could catch them earlier?

[jamadden]

I found a 32-bit Python to test on, so hopefully PR #23 should really do it.

[mgedmin]

The failures go to zope-tests@zope.org, and then a daily summary is sent to zope-dev@zope.org (but it's basically delayed by a whole day). I use this script to analyze the daily summaries in a non-painful way.

[jamadden]

Thank you.