Andrew Gallant ab72269b23 Add ASCII word boundaries to the lazy DFA. ...

In other words, `\b` in a `bytes::Regex` can now be used in the DFA.
This leads to a big performance boost:

```
sherlock::word_ending_n                  115,465,261 (5 MB/s)  3,038,621 (195 MB/s)    -112,426,640  -97.37%
```

Unfortunately, Unicode word boundaries continue to elude the DFA. This
state of affairs is lamentable, but after a lot of thought, I've
concluded there are only two ways to speed up Unicode word boundaries:

1. Come up with a hairbrained scheme to add multi-byte look-behind/ahead
   to the lazy DFA. (The theory says it's possible. Figuring out how to
   do this without combinatorial state explosion is not within my grasp
   at the moment.)
2. Build a second lazy DFA with transitions on Unicode codepoints
   instead of bytes. (The looming inevitability of this makes me queasy
   for a number of reasons.)

To ameliorate this state of affairs, it is now possible to disable
Unicode support in `Regex::new` with `(?-u)`. In other words, one can
now use an ASCII word boundary with `(?-u:\b)`.

Disabling Unicode support does not violate any invariants around UTF-8.
In particular, if the regular expression could lead to a match of
invalid UTF-8, then the parser will return an error. (This only happens
for `Regex::new`. `bytes::Regex::new` still of course allows matching
arbitrary bytes.)

Finally, a new `PERFORMANCE.md` guide was written.

2016-04-08 22:58:10 -04:00

..

src

Add ASCII word boundaries to the lazy DFA.

2016-04-08 22:58:10 -04:00

Cargo.toml

regex-syntax 0.3.1

2016-03-28 16:32:36 -04:00