log2(0) panicked at 'attempt to subtract with overflow', parquet/src/util/bit_util.rs:148:5

[HaoYang670]

Describe the bug
https://github.com/apache/arrow-rs/blob/master/parquet/src/util/bit_util.rs#L142-L155

To Reproduce
assert_eq!(log2(0), 0);

Expected behavior
Alter 1. remove it and replaced by num_required_bits
Alter 2: return Some(n) when input > 0 and None when input == 0;
Alter 3: use the std function core::num::log2 which has not been stable yet. https://doc.rust-lang.org/std/primitive.u64.html#method.log2
Alter 4: panic when input == 0 (follow the behavior of num::core::log2)
Alter 5: return 0 when input == 0 (follow the behavior of num::core::log2)

Additional context
Add any other context about the problem here.

[jhorstmann]

Since this seems to be only used for getting the width of bitpacked data, a faster way to calculate that is using the leading_zeros function. I contributed a similar improvements to parquet2 some time ago: https://github.com/jorgecarleitao/parquet2/blob/34aac65e433cff8bdd9fb07147cb551584a5415c/src/read/levels.rs#L3

[HaoYang670]

Thank you @jhorstmann.
Agree with you on using leading_zeros, and this is also how the std library implements:

                #[inline]
                pub const fn log2(self) -> u32 {
                    Self::BITS - 1 - self.leading_zeros()
                }

But the corner case is what should the behavior of log2(0) be.

[HaoYang670]

Ha! Maybe we could remove this function directly because we already have num_required_bits which provide same functionality.
https://github.com/apache/arrow-rs/blob/master/parquet/src/util/bit_util.rs#L180-L189