Antoine, thanks for the patch.  This looks like a reasonable solution that is fast and prevents segfaults.

The error message should say more than just that the size changed. it should also note that the heap may no longer be in the heap condition since it was interrupted in the middle of an update.

I'm also wondering whether a stronger check is needed than detecting a length change (perhaps something akin to the logic used in Py_ReprEnter and Py_ReprLeave).