Feature Request (globset): Expose `MatchStrategy` about ripgrep

[BurntSushi]

No, it's an implementation detail. It's not totally clear to me why your use case needs it to be public.

[sxyazi]

I want to know what kind of glob the user has set, such as whether it is just a simple extension glob. If it is, I can go through my own optimized processing logic without compiling a regular expression for it and performing regex matching, and only need to compare a small part of the path.

So I need to detect the type of glob the user has set and decide on a strategy, and I found that globset has implemented this very well just not make it public, of course I could refer to globset and re-implement it from scratch, but I think that's unnecessary workload.

[sxyazi]

Add more details:

After reading the source code of globset, I found that only GlobSet uses this strategy optimization during matching, while single glob matching (GlobMatcher) does not (please correct me if I'm wrong).

However, GlobSet::matches() matches all glob rules and then returns all indexes. For my needs (only needing to get the first matching icon), this is an unnecessary additional overhead because it could have ended early.

So what I need is a combination of strategy matching and early termination, which globset cannot provide since it uses strats: Vec<GlobSetMatchStrategy> to save each glob grouped according to the strategy, as the order between globs is no longer available, which is why it must calculate all the indexes and reorder them:

ripgrep/crates/globset/src/lib.rs

Lines 406 to 410 in d922b7a

	for strat in &self.strats {
	strat.matches_into(path, into);
	}
	into.sort();
	into.dedup();

Therefore, I think if the MatchStrategy is made public, I can implement it myself.

[BurntSushi]

globset already avoids compiling a regex in some cases. Perhaps more cases could be added.

Sorry, but I don't want to expose internals like what you're requesting. I would be open to adding optimizations though.

[sxyazi]

OK no problem.

globset already avoids compiling a regex in some cases. Perhaps more cases could be added.

I couldn't find the relevant code, could you give me some hints? I'll see if I can add it - I'm interested in optimizing extension globs like *.jpg in GlobMatcher.

[BurntSushi]

I'm confused because you are literally asking about the very thing that does what you want. It even has an optimization path for extensions.

ripgrep/crates/globset/src/glob.rs

Lines 155 to 177 in d922b7a

	match self.strategy {
	MatchStrategy::Literal(ref lit) => lit.as_bytes() == byte_path,
	MatchStrategy::BasenameLiteral(ref lit) => {
	lit.as_bytes() == &*candidate.basename
	}
	MatchStrategy::Extension(ref ext) => {
	ext.as_bytes() == &*candidate.ext
	}
	MatchStrategy::Prefix(ref pre) => {
	starts_with(pre.as_bytes(), byte_path)
	}
	MatchStrategy::Suffix { ref suffix, component } => {
	if component && byte_path == &suffix.as_bytes()[1..] {
	return true;
	}
	ends_with(suffix.as_bytes(), byte_path)
	}
	MatchStrategy::RequiredExtension(ref ext) => {
	let ext = ext.as_bytes();
	&*candidate.ext == ext && self.re.is_match(byte_path)
	}
	MatchStrategy::Regex => self.re.is_match(byte_path),
	}

It does not appear to skip compiling the regex, though, it does not use it for matching. I'd be open to adjustments that skip the regex compilation where possible. I can't remember whether that will be hard or not though. It could be a big refactor. But maybe not.