gh-91117: Ensure integer mod and pow operations use cached small ints

[sweeneyde]

https://bugs.python.org/issue46961

[sweeneyde]

I found the call graph of function names a lot to juggle, but the relevant call graphs are like this:

long_divmod:
    l_divmod:
        fast_floor_div
        fast_mod
        long_divrem (doesn't check for small remainders):
            divrem1 (returns fresh longs)
            x_divrem (returns fresh longs)

long_mod:
    l_mod:
        fast_mod
        long_rem (doesn't check for small remainders):
            rem1
            x_divrem (returns fresh longs)

long_pow:
    l_mod:
        (same as above)
 

As I noted in the issue, x_divrem always returns new freshly-allocated ints. Its callers are like this:

    - long_divrem
        - uses maybe_small_long on quotient
        - doesn't check if remainder is small <---- oops
    - long_rem
        - throws away quotient
        - doesn't check if remainder is small <---- oops
    - long_true_divide
        - modifies the quotient (!!!)
        - throws away remainder

[mdickinson]

@sweeneyde Sorry Dennis. This is on my list of things to review, but I'm not likely to get to it for a few days.

[mdickinson]

LGTM!

[mdickinson]

It surprised me initially that it was okay to call maybe_small_long on a value that might be NULL, but I see that other parts of Objects/longobject.c also do that (and it makes sense from a convenience perspective).

[ghost]

Commit authors are required to sign the Contributor License Agreement.

[sweeneyde]

Thanks for the review! Looks like we should wait for Blurb-It to sign its CLA ;)

[mdickinson]

Looks like we should wait for Blurb-It to sign its CLA ;)

I think we can reasonably bypass that. Feel free to merge!