Two concerns before this lands.

The 1000000 cap is arbitrary and rejects computable inputs, e.g. gmp_pow(2, 2000000) is a ~250 KB integer this now throws on. A fixed exponent limit was added once before and reverted as too restrictive (GH-16870, #16973), so a magic constant will likely draw the same objection.

It also caps only the exponent, not the result size. The GMP abort is driven by exp * bitlen(base), so a large enough base still reaches __gmp_exception under this cap, e.g. gmp_pow(gmp_init(str_repeat('9', 200000)), 1000000). Bounding against the projected size (exp * mpz_sizeinbase(base, 2) versus GMP's limit) would cover both cases and drop the magic number.

The GMP abort is driven by exp * bitlen(base), so a large enough base still reaches __gmp_exception under this cap, e.g. gmp_pow(gmp_init(str_repeat('9', 200000)), 1000000). Bounding against the projected size (exp * mpz_sizeinbase(base, 2) versus GMP's limit) would cover both cases and drop the magic number.

Thank you. Good points — agreed, the 1,000,000 cap is arbitrary and can reject valid cases like gmp_pow(2, 2000000).

I think the issue is caused by this calculation: mpz/n_pow_ui.c:371

ralloc = (bsize * GMP_NUMB_BITS - cnt + GMP_NAIL_BITS) * e / GMP_NUMB_BITS + 5;