Continuing the discussion from Issue 3552
(http://bugs.python.org/issue3552).
r65676 makes changes to Modules/md5module.c and Modules/sha1module.c, to
allow them to read mutable buffers.
There's a segfault in sha1module if given 0 arguments. eg:
>>> import _sha1
>>> _sha1.sha1()
Segmentation fault
Docs here suggest this should be OK:
http://docs.python.org/dev/3.0/library/hashlib.html
This crashes on the Lib/test/test_hmac.py test case, but apparently
(according to Margin on issue 3552) none of the build bots see it
because they use libopenssl and completely bypass the _md5 and _sha1
modules. Also there are no direct test cases for either of these modules.
This is because new code in r65676 doesn't initialise a pointer to NULL.
Fixed in patch (as well as replaced tab with spaces for consistency, in
both modules).
I strongly recommend that a) A "build bot" be made to use _md5 and _sha1
instead of OpenSSL (or they aren't running that code at all), AND/OR b)
Direct test cases be written for _md5 and _sha1.
Commit log:
Fixed crash on _sha1.sha1(), with no arguments, due to not initialising
pointer.
Normalised indentation in md5module.c and sha1module.c.
|