add anchored search APIs

[Alphare]

In Mercurial, we have to use ^(?:<patterns>) where <patterns> is the culmination of all patterns (transformed to regex) in the user's .hgignore to remove the additional .* on each end. The user could input a regex like )(, which while invalid on its own would work with the workaround and would create a useless capturing group.

Adding an option to RegexBuilder and its bytes cousin seems like a good solution to me.

[BurntSushi]

OK, so I have some varied thoughts on this.

I think at a high level, adding an option as described here is probably the wrong path to take. In particular, your problem doesn't really have anything to do with ^ specifically, but rather, it's a problem of composition. ripgrep actually does similar things with regex composition in order to implement its -w/--word-regexp flag, but rg ')(' -w correctly exits with a syntax error. This is because it first attempts to parse the regex as given before trying to compose them. If you're doing regex composition, then I think this is the only correct way to do it. More specifically, you could depend on regex-syntax to run just the parser to check for syntax validity. It's a little extra work, but should overall be pretty cheap compared to the entire regex compilation process.

With that said, an "anchored search" is indeed kind of a special case. I think my plan at the moment is not to surface this as a compile-time option, but rather a search-time option. That is, perhaps in addition to Regex::find there will also be Regex::find_anchored (or whatever name). But that needs at least some API design work and won't happen until at least #656 is done.

[Alphare]

You're right that checking the pattern first with regex-syntax should be pretty inconsequential in terms of runtime compared to building the DFA and even more so compared to the rest of the program. I have to thank whoever decided to split regex in modular crates to make this so easy. ;)

We will be using this "workaround" (if that's really the term) until Regex::find_anchored becomes part of the API, thanks.

[BurntSushi]

No problem. And yeah, it's kind of a work-around for this specific case, but for general composition, I think it's right.

It is plausible that some kind of API for this should/could be exposed in regex proper. Maybe not full syntax parsing, but a, for example, parse_regex(&str) -> Result<(), Error> that just checked whether the regex was valid or not without compiling I think would be sufficient for composition. Then folks wouldn't need to depend on regex-syntax explicitly. (Which, while convenient, is still primarily supposed to be an implementation detail of regex.)

to building the DFA

Just to make sure your mental model is right here, the regex crate currently never builds a full DFA ahead of time. It builds an NFA first, and depending on which matching engine is selected, will either execute the search directly with the NFA or will build the DFA lazily one state at a time during a search. This is the same execution model as RE2.

(In the future, I expect there will be some cases where building the DFA ahead of time is done, but only when doing so would be very cheap and use very little space.)

[BurntSushi]

I think once #656 lands, it will be possible to achieve this using regex-automata's "meta" regex engine. It will support this sort of flexibility with a richer set of search options.

I'm not sure if it will ever make it into regex proper through unfortunately, since it would seem to me to require duplicating a lot of the methods.

So for now, I think I'm going to request that folks who need this try out the meta regex engine once regex-automata 0.3 is out. If you run into troubles there, then please file an issue.