./python -m test -v test_xml_etree
...
test_getslice_negative_steps (test.test_xml_etree.ElementSlicingTest) ...
python: Objects/sliceobject.c:176: _PySlice_AdjustIndices: Assertion `step >= -PY_SSIZE_T_MAX' failed.
...
Traceback (most recent call first):
  File "/home/haypo/prog/python/2.7/Lib/xml/etree/ElementTree.py", line 266, in __getitem__
    return self._children[index]
  File "/home/haypo/prog/python/2.7/Lib/test/test_xml_etree.py", line 2305, in test_getslice_negative_steps
    self.assertEqual(self._elem_tags(e[3::-sys.maxsize-1]), ['a3'])

See issues: bpo-30074, bpo-27867.

Attached PR fixes test_xml_etree.

Thank you Victor for noticing and fixing this crash!

New changeset f6a3133972378205baaa6a854d46170d04a2db67 by Victor Stinner in branch '2.7':
bpo-30255: Clip step in _PySlice_Unpack() (#1429)
https://github.com/python/cpython/commit/f6a3133972378205baaa6a854d46170d04a2db67

I pushed the change to try to fix buildbots, but I'm not sure that it's complete.

The change announces "... clip the step to [-PY_SSIZE_T_MAX,
PY_SSIZE_T_MAX] ...", but there is no explicit to clip the maximum to PY_SSIZE_T_MAX. Maybe there is a bug on this side too! The code should be reviewed one more time.

> Thank you Victor for noticing and fixing this crash!

Something is *very* strange: we have +20 buildbots, but I didn't see this assertion failure on any buildbot. Instead, I noticed that regrtest runs less tests than announced: see issue #30254.

It's very strange because I expect either a crash: regrtest stops with no more output, or a report of the crash. But no, regrtest displays the summary as everything was fine...

> It's very strange because I expect either a crash: regrtest stops with no more output, or a report of the crash. But no, regrtest displays the summary as everything was fine...

It was a bug in regrtest: now fixed by bpo-30258.

> The change announces "... clip the step to [-PY_SSIZE_T_MAX, PY_SSIZE_T_MAX] ...", but there is no explicit to clip the maximum to PY_SSIZE_T_MAX. Maybe there is a bug on this side too! The code should be reviewed one more time.

Ok, I checked: _PySlice_Unpack() uses _PyEval_SliceIndex() to convert the step Python object into a C Py_ssize_t, so the code is already correct: it does "clip the step to [-PY_SSIZE_T_MAX, PY_SSIZE_T_MAX]" as announced ;-) I close the issue.

Thanks Serhiy for your work on fixing corner cases related to slices ;-)