Is it possible to build a regex::Regex from a regex_automata::meta::Regex?

[xmo-odoo]

I'm using regex_syntax to work with HIR then converting the HIR to a regex_automata for matching proper, however to expose to users regex_automata might be a bit low-level and involved (especially when it comes to captures), so I wanted to know if there was an easy way to convert into a Regex with all the convenient high-level helpers, or if I'd have to either expose the lower level APIs or create my own convenience wrapper.

[BurntSushi]

Good question. This is similar to questions folks have asked in the past (before regex-automata existed) like, "If I have an Hir from regex-syntax, can I use that to directly build a regex::Regex?" The answer to that is basically the same answer to your question: definitively and intentionally not. The only way for a conversion like the one you're asking for to exist is if regex-automata were a public dependency of regex. That in turn would mean that every time regex-automata had a semver incompatible release, regex would also need a semver compatible release. This is not something I can abide, because it would imply that regex-automata could not evolve independently of regex. That is, it would inhibit its primary goal. From the docs (emphasis mine):

The primary goal of this crate is to serve as an implementation detail for the regex crate. A secondary goal is to make its internals available for use by others.

With that said, not all hope is lose here because there is a stable interface between regex-syntax and regex: the concrete syntax. If you're working with Hir, then you can always call hir.to_string() and pass it to Regex::new. ripgrep used this for years before it switched over to regex-automata for its matching. (ripgrep doesn't even use regex directly.) In particular, the conversion from Hir to a string, and then to a regex via Regex::new is guaranteed to work.

It does sound wasteful, and technically, it is. But necessarily so given the constraints above. With that said, it is probably not as wasteful as you might think. In particular, the parsing and creation of an Hir would be repeated, but this is one of the least expensive parts of Regex::new. It is much more expensive to build the internal finite state machine data structures, and that is not being repeated by the following process: parse -> HIR -> modify HIR -> to string -> Regex::new.

[BurntSushi]

Also, the only reason why you can build a meta::Regex from an Hir is because regex-syntax is a public dependency of regex-automata. This is okay because regex-syntax has been baking for a long time and I generally expect it to evolve more slowly than regex-automata.

[xmo-odoo]

Thanks, that seems fair. I don't actually modify the HIR (only parse / datamine it), so I can convert all the regex-automata bit to straight regex, and re-parse the source expression, with some sort of bridge for the flags I might be using so they're used for both HIR parsing and regex.