how to wrap a large alternation of literals in word boundary assertions and have it be fast?

[Guillermogsjc]

Hi, to match efficiently large amounts of alternations, I guess it is interesting to trigger aho_corasick variant here

regex/src/exec.rs

Line 91 in 9ca3099

/// An Aho-Corasick automaton with leftmost-first match semantics.

regarding doc:

/// This is only set when the entire regex is a simple unanchored
/// alternation of literals. We could probably use it more circumstances,
/// but this is already hacky enough in this architecture.

The question is: is there any way to use word boundaries in such a way this expression is highly optimized for a thing like this?

r"\b(a|... #massive ammount of literal alternations here# ...|z)\b"

or with (?-u:\b) instead of \b .

And... regarding PERFORMANCE documentation here

there is no problem with using non-greedy matching or having lots of alternations in your regex

this previously stated regex would be in the set of "no problem" ?

Thanks

[BurntSushi]

I've turned this into a feature request in #891.

And... regarding PERFORMANCE documentation here

there is no problem with using non-greedy matching or having lots of alternations in your regex

this previously stated regex would be in the set of "no problem" ?

Yes. But that doesn't mean it gets the Aho-Corasick optimization. The section you're quoting is really written relative to backtracking engines. Namely, a backtracking engine will (absent other optimizations) try each alternate one after the other. This will wind up being slower than even the slowest regex engine in this crate (the PikeVM).

I am being somewhat hand wavy here. Being precise here without concrete examples in front of us is difficult.