Prefer Vec<u8>/[u8] over Vec<char>/[char]

[TianyiShi2001]

When creating a Vec<char> from a string, s.chars().collect() is the standard practice. The .chars() method works on any UTF-8 encoded string and it takes time to find the boundary for each character. However, we know that PDB files contain ASCII characters only, therefore it will be faster to transmute a string directly to bytes using as_bytes() and use vectors/arrays/slices of bytes instead of chars throughout the crate. This practice is used, for example in rust-bio and seq-io

[douweschulte]

I totally agree.

[TianyiShi2001]

Doing so requires that when we parse a PDB file we assume it's all ASCII, is it OK? (the checks that are run when modifying the structure are preserved)

[DocKDE]

Out of curiosity: what happened to this suggestion? I'm asking because I took another look during profiling and a decent chunk of the time pdbtbx needs for parsing is currently used up by collecting chars into Strings. I'm not sure if this would be impacted by a switch to bytes but since PDB files only contain ASCII characters the crate could benefit from such a change, no?

[douweschulte]

The progress stalled. But changing to u8 will not necessarily decrease the time spent on collecting to Vecs. The most benefit will be in the use of the more specialised u8 based ascii functions over UTF8. If most time is spent on collecting to Vecs I think it would be more beneficial to find out why and if the collection can be removed/sped up, maybe by providing the final size of the collection. If you want feel free to work on this.