Switch away from std::regex to RE2

[Zaczero]

Is your feature request related to a problem?

I've been interested in improving the Nix performance.

Proposed solution

Switch away from std::regex to RE2.

Alternative solutions

• boost::regex (already installed)
• Do nothing.

Additional context

I'm not C++ kind of guy.

I've been interested in improving the Nix performance and I noticed that std::regex is used quite frequently. When benchmarking I saw it puts quite noticeable amount of overhead whenever used. I confirmed my suspicion by searching around online e.g., here. In the past I had pleasant experience with working with RE2 because it's generally faster and safer to use. I suspect it may be a good fit for Nix.

Checklist

• checked latest Nix manual (source)
• checked open feature issues and pull requests for possible duplicates

---

Add 👍 to issues you find important.

[xokdvium]

See some discussions around #7762. Long story short: switching to either boost regex or re2 would undoubtedly break things. There's apparently even a case in nixpkgs which depends in a backwards-incompatible way on a libstdc++ bug :(

[Zaczero]

How about we keep it internal? For cases like these: #14863

[xokdvium]

For some of those cases I did start switching away to boost regex. It's much more performant than the standard library and doesn't require pulling in another dependency.

[Zaczero]

I believe performance and safety benefits justify RE2 if it doesn't introduce runtime incompatibilities. It makes it impossible to write unsafe regex and has a linear time matching guarantee (which matches the spirit of Nix: more predictable and deterministic). But that's just my opinion and I'm very new to Nix source.

[xokdvium]

Maybe let's first try boost.regex when we are able to identify bottlenecks and see from there? It would be easier to justify in that case. From my experience boost.regex is more than passable for what we use it for.