Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> writes:
> The problem then arises in the final carry propagation pass. During
> this phase of the computation, the carry from one digit (which can be
> a shade under INT_MAX / NBASE) is added to the next digit, and that's
> where the overflow happens.
Nice catch! I think the comment could use a little more work, but I'll
adjust it and push.
regards, tom lane