GMP error handling

[Kevin Ryde]

nisse at lysator.liu.se (Niels Möller) writes:
>
> I usually prefer return codes from the failing function, but an error
> callback on out-of-memory sounds like it would at least be an
> improvement over hacks like longjmp:ing out of the memory allocation
> functions.

I don't think the memory stuff will change.  The alloc/realloc
functions must return a valid block, or not return at all.