Message11241

I have been looking at the test_sort failure and now have managed to create a simple (ish) way of reproducing the failure test_sort is having. After spending quite a while debugging it one issue is that the test makes lots of use of randomness to shuffle up list and to make exceptions occur randomly, this isn't very helpful when trying to understand the issue.

So I have made a script which reproduces one of these failures reliably which is attached. Its based on an observed failure while running the real test.

Essentially the problem is after calling sort() the returned list now contains duplicated elements and other elements have been removed, this does seem like quite a bad issue and I still don't think I understand why this is happening, but though a way to reproduce it might be helpful for others to investigate.

I made the script also run in cpython however the order of the comparisons there is different so it never actually triggers the RuntimeError, but it does work as expected.

Not sure if this issue should be blocking for 2.7.1 it does seem bad?