Faster deref of inline atoms

[overlookmotel]

Deref of an atom (e.g. atom.to_string()) which stores the string inline performs slicing with &src[..len]. len is guaranteed to be within bounds, as an inline atom is always 7 bytes or less, so this can be substituted with src.get_unchecked(..len).

Removing this bounds check improves performance of deref-ing an inline atom by approx 30%.

There's a lot of noise in the benchmarks, but it also seems to slightly improve performance of deref-ing static and dynamic atoms too, presumably because it reduces the size of the deref() function.

NB There's no gain to making the same change in From<Cow<'a, str>> for Atom<Static> because the compiler already deduces no bounds check is required due to the if len <= MAX_INLINE_LEN conditional.

[overlookmotel]

@nicoburns Thanks for reviewing. Are you a maintainer of this project and able to merge?

[nicoburns]

@nicoburns Thanks for reviewing. Are you a maintainer of this project and able to merge?

I am, but I'm new to working with this crate, so I'm going to seek feedback before I go ahead and merge. I will look to push this forwards though :) See: https://servo.zulipchat.com/#narrow/stream/263398-general/topic/The.20.60string_cache.60.20crate

[overlookmotel]

@nicoburns Amazing! Would be really great if this crate received some attention, and really appreciate it that you're willing to do so.

[overlookmotel]

Thanks very much for merging!