homepage

This appears to be 2.7-only:
$ ./python -m test.regrtest -M5G -v test_hashlib
== CPython 2.7.3+ (2.7:086afe7b61f5, May 23 2012, 15:15:34) [GCC 4.5.2]
== Linux-2.6.38.8-desktop-10.mga-x86_64-with-mandrake-1-Official little-endian
== /home/antoine/cpython/27/build/test_python_6042
Testing with flags: sys.flags(debug=0, py3k_warning=0, division_warning=0, division_new=0, inspect=0, interactive=0, optimize=0, dont_write_bytecode=0, no_user_site=0, no_site=0, ignore_environment=0, tabcheck=0, verbose=0, unicode=0, bytes_warning=0, hash_randomization=0)
test_hashlib
test_algorithms_attribute (test.test_hashlib.HashLibTestCase) ... ok
test_case_md5_0 (test.test_hashlib.HashLibTestCase) ... ok
test_case_md5_1 (test.test_hashlib.HashLibTestCase) ... ok
test_case_md5_2 (test.test_hashlib.HashLibTestCase) ... ok
test_case_md5_huge (test.test_hashlib.HashLibTestCase) ... python: /home/antoine/cpython/27/Modules/md5module.c:276: MD5_new: Assertion `(Py_ssize_t)(unsigned int)(view.len) == (view.len)' failed.
Abandon

Here is a patch.

Does this affect other hash modules?. Why is this not affecting python 3?
Patch looks good.

> Does this affect other hash modules?
I don't know, only md5 seems to have tests for large data.
> . Why is this not affecting python 3?
The _md5 module was apparently rewritten in Python 3.

I can't reproduce this issue in my 64 bit machines, neither in Solaris neither in Ubuntu. I guess the assertion can be fooled by compiler optimizacions.
As a consequence, I can't check other hashes functions.
Antoine, can you reproduce it doing "md5.new("A"*(2**32+5)).hexdigest()"?. That is what test HUGE does.

> I can't reproduce this issue in my 64 bit machines, neither in Solaris
> neither in Ubuntu. I guess the assertion can be fooled by compiler
> optimizacions.
You should compile in debug mode.

sha1 fails the same way. Same error. Just clone the test to show it.
Please, correct sha1 too and add a test for it :).
sha224, sha256, sha384 and sha512 seems OK.

sha224, sha256, sha384 and sha512 are not failing because they are missing the "Py_SAFE_DOWNCAST" safety net completely.
So I would tell that we have an issue here :).

Same can be said about Python 3 hash modules: they are not using the sanity check, so they work.
Maybe the real question should be if the sanity check really makes sense at all. If not, remove everywhere (the calculated md5 with no checks looks correct, after all). If it makes sense, mark this as "python 3.2 and 3.3" and patch everywhere :).
Do you agree, Antoine?

> sha1 fails the same way. Same error. Just clone the test to show it.
> 
> Please, correct sha1 too and add a test for it :).
Well, do you want to provide an updated patch?

New changeset 290d970c011d by Antoine Pitrou in branch '2.7':
Issue #14888: Fix misbehaviour of the _md5 module when called on data larger than 2**32 bytes.
http://hg.python.org/cpython/rev/290d970c011d 

I've now pushed the fix. Jesus, if you want to propose a test and patch for the _sha1 issue, please open a separate issue. Thanks!