Some internal C functions represent the number of bits in the Python integer as size_t or Py_ssize_t. It is fine on 64-bit platforms, where you need exbibytes of memory to get an overflow error. But on 32-bit platform you can create an integer objects that has a size of just 0.5 GiB.

This problem can be solved if always use 64-bit integers (uint64_t or int64_t) for bit counts. We can even introduce a hard limit for the range of integers in CPython (to $2^{2^{64}-1}$ or $2^{2^{63}-1}$) and remove the possibility of overflow error in _PyLong_NumBits() and _PyLong_Frexp(). No existing 64-bit platform supports such large address space, and even if they support, it would take years to create a single integer object of such size (just to fill memory).

Linked PRs

• gh-121485: Always use 64-bit integers for integers bits count #121486