Fix potential underflow in bigint_to_scientific

[Jay Lorch (jaylorch)]

Now that Verus has some support for format!, I was able to start checking functions that use it, including bigint_to_scientific. This uncovered a potential panic in that function. If usize is 64 bits (as it is on many platforms), then digits.len() can be 2^31 or greater. If it's equal to 2^31, or congruent to 2^31 modulo 2^32, then casting to an i32 will produce i32::MIN, and then subtracting 1 will panic due to underflow. This PR prevents that in a pretty straightforward way: It just doesn't bother casting to an i32.

[Copilot]

Pull request overview

This pull request adjusts bigint_to_scientific in src/number.rs to avoid a potential overflow/underflow panic caused by casting digits.len() from usize to i32 before subtracting 1.

Changes:

• Replace digits.len() as i32 - 1 with digits.len() - 1 to avoid overflow-sensitive casting.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Jay Lorch (jaylorch)]

Your suggestion leads to incorrect output, giving the wrong exponent. I don't understand how it fixes anything.

[Anand Krishnamoorthi (anakrish)]

Thanks for identifying and fixing this! It would be great to prove absence of panics.