I am doing blockchain indexing and when analyzing perf output of my indexer (which is quite CPU-bound at this point), I can see 10% of CPU time sitting in hex_bytes. I query blocks from the node, which get decoded from hex response to Vec<u8> and then to Block via consensus_decode. All of this happens in rust-bitcoincore-rpc https://github.com/rust-bitcoin/rust-bitcoincore-rpc/blob/master/client/src/queryable.rs#L23

Looking at:

https://github.com/rust-bitcoin/rust-bitcoin/blob/master/src/util/misc.rs#L24

I see the following room for improvement:

• output vector's capacity could be preallocated (should be typically 1/2 of the string length)
• chars is for handling Unicode chars, so I suspect there's quite a bit of overhead there. We can assume that the hex string is ASCII, and just convert it to &[u8] directly via as_bytes() and work on that; unicode characters are > 128 so it all should be fine;
• the whole iterator-based approach seems suspicious, performance-wise; pair, fold; who knows how it gets optimized; a simpler, imperative code would be both more readable, and maybe faster