test_structmembers fails on s390x (bigendian 64-bit): int/Py_ssize_t issue

[davidmalcolm]

BPO	9960
Nosy	@mdickinson, @pitrou, @davidmalcolm, @serhiy-storchaka
Files	python-test_structmembers.patch fix-test_structmember-on-64bit-bigendian.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2018-12-12.09:39:24.764>
created_at = <Date 2010-09-27.15:11:33.902>
labels = ['easy', 'type-bug', 'tests']
title = 'test_structmembers fails on s390x (bigendian 64-bit): int/Py_ssize_t issue'
updated_at = <Date 2018-12-12.09:39:24.761>
user = 'https://github.com/davidmalcolm'

bugs.python.org fields:

activity = <Date 2018-12-12.09:39:24.761>
actor = 'serhiy.storchaka'
assignee = 'none'
closed = True
closed_date = <Date 2018-12-12.09:39:24.764>
closer = 'serhiy.storchaka'
components = ['Tests']
creation = <Date 2010-09-27.15:11:33.902>
creator = 'dmalcolm'
dependencies = []
files = ['19029', '25199']
hgrepos = []
issue_num = 9960
keywords = ['patch', 'easy']
message_count = 6.0
messages = ['117448', '117951', '128576', '158177', '221826', '331685']
nosy_count = 5.0
nosy_names = ['mark.dickinson', 'pitrou', 'dmalcolm', 'BreamoreBoy', 'serhiy.storchaka']
pr_nums = []
priority = 'normal'
resolution = 'out of date'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue9960'
versions = ['Python 2.7']

[davidmalcolm]

test test_structmembers crashed -- <type 'exceptions.ValueError'>:
string too long
Traceback (most recent call last):
  File "/builddir/build/BUILD/Python-2.7/Lib/test/regrtest.py", line 863, in runtest_inner
    the_package = __import__(abstest, globals(), locals(), [])
  File "/builddir/build/BUILD/Python-2.7/Lib/test/test_structmembers.py", line 12, in <module>
    9.99999, 10.1010101010, "hi")
ValueError: string too long

_testcapimodule.c: test_structmembers_new's fmt has:
"s#"
and these args:
&s, &string_len

for grabbing this data:
    const char *s = NULL;
    Py_ssize_t string_len = 0;

However, the module doesn't define PY_SSIZE_T_CLEAN, which leads to &string_len being treated as an (int*) rather than a (Py_ssize_t*) and thus written to with just the first 32-bits of the size, rather than the full size.

The PyArgs_ call without PY_SSIZE_T_CLEAN writes the size of the string (2) through (int*)&string_len, writing 0x00000002 to the first 4 bytes, setting the 64-bit "string_len" value to 0x0000000200000000 i.e. 2^34, wildly too large for the iirc 5 byte buffer.

Confirmed in gdb:
(gdb) p string_len
$2 = 8589934592

(gdb) p /x string_len
$3 = 0x200000000

Am attaching a patch (against 2.7 maintenance branch) which defines PY_SSIZE_T_CLEAN; doing so requires updating another int to be a Py_ssize_t in that module, within test_u_code.

http://docs.python.org/c-api/arg.html lists "u# (Unicode) [Py_UNICODE *, int]" and has no reference to the effect of PY_SSIZE_T_CLEAN on the "u#" format specifier. My reading of Python/getargs.c is that this macro does affect "u#" in a manner analogous to "s#"; does this documentation need clarifying?

[pitrou]

Is this 2.7-specific? Otherwise, it would be better to provide a patch for 3.2 first, and then svnmerge to other branches.

My reading of Python/getargs.c is that this macro does affect "u#" in a > manner analogous to "s#"; does this documentation need clarifying?

Probably. Same for "es#", "et#" and "t#" perhaps?

[davidmalcolm]

FWIW, this regressed in 2.6.6 relative to 2.6.5, due to r79646 adding the T_STRING_INPLACE test code to the 2.6 branch.

Note to self: https://bugzilla.redhat.com/show_bug.cgi?id=677392

[davidmalcolm]

The originally attached patch is no good for the the 2.* branch, as it appears that _testcapimodule.c will not become "ssize_t" safe in Python 2.*; see e.g.:
http://hg.python.org/cpython/rev/3ecddf168f1f

Am attaching a revised patch that I'm applying downstream in Fedora's builds of 2.7.3

[BreamoreBoy]

I believe that the change in the later patch has already been applied to the 2.7 branch. If I'm correct can we close this please.

[serhiy-storchaka]

This was fixed in bpo-17928.