> Maybe we could consider adding an extra field to a PyLong giving its
> 'small_int' value for small values, and some flag value for non-small
> longs.  An extra field wouldn't actually enlarge the size of a PyLong
> for small values---on a 64-bit machine, a value like 23L takes 28
> bytes, for which 32 bytes will actually be allocated (since Python
> always allocates in multiples of 8 bytes, I believe).

I actually had a patch for that. It declared a ob_digit[2] array instead
of ob_digit[1], and ob_digit[1] contained the small int. But the patch
still used the pointer comparison approach for PyLong_IS_SMALL_INT,
because I think it's faster (just two comparisons). So there didn't seem
to much point.

Also, the pointer addition trick for addition (see BINARY_ADD) is
probably faster than the more intuitive method.