regex-dna with &[u8] matching

[llogiq]

I have a gist where I use regex::bytes instead of plain regex. I'd have thought that it should be faster (because we can get rid of the UTF-8 check), but in fact it is slower. @BurntSushi there's probably some reason for that, just a heads-up; if the bytes variant should get faster we can retry the measurements.

On my system:

variant	wall	user	sys
UTF-8	1.52	3.94	0.33
bytes	2.23	4.50	0.35

[BurntSushi]

Yeah, I noticed this a few weeks ago and though it was curious. I actually can't think of any obvious reason why it's slower. My (more extensive) benchmark suite says Regex::new and bytes::Regex::new have comparable performance. I'll look into it.

Another thing you could try is from_utf8_unchecked.

[BurntSushi]

Ahahahahaha. This is because replace_all in Regex uses push_str, which should be a memcpy. On the other hand, bytes::Regex uses extend, which seems to be slower, and therefore I'm guessing doesn't get optimized down to a memcpy. Nice.

[BurntSushi]

OK, once I fixed that in regex, the best of 3 for bytes::Regex was 0.59s and for Regex it was 0.61s. I figured it would be that small. It's barely perceptible and probably in the noise because I not-so-infrequently see both of them get up to 0.7s.

[llogiq]

Ok then I'm going to wait for a new Regex release so I can submit it without complicating our makefile. Thanks @BurntSushi!