FixedLenByteArray max_precision integer overflow

[evanrichter]

fuzzing found a multiplication overflow here:

parquet2/src/schema/types/spec.rs

Line 44 in 7caafa4

(2f64.powi(8 * (*length as i32) - 1) - 1f64).log10().floor() as usize;

so I changed it to a checked_mul and also realized the subtract 1 could underflow also. the fuzzer didn't confirm, but the rust compiler did:

error: this arithmetic operation will overflow
 --> src/main.rs:2:20
  |
2 |     println!("{}", (i32::MIN / 8) * 8 - 1);
  |                    ^^^^^^^^^^^^^^^^^^^^^^ attempt to compute `i32::MIN - 1_i32`, which would overflow

[jorgecarleitao]

Thanks, @evanrichter 🙇