codestats 0.6.0

use std::{cell::RefCell, sync::LazyLock};

use globset::{GlobBuilder, GlobSet, GlobSetBuilder};

use super::super::data::{LANGUAGES, Language};

pub(super) struct LanguageGlobs {
	pub(super) set: GlobSet,
	pub(super) pattern_lang_indexes: Vec<usize>,
}

pub(super) static LANGUAGE_GLOBSET: LazyLock<LanguageGlobs> = LazyLock::new(|| {
	let mut builder = GlobSetBuilder::new();
	let mut pattern_lang_indexes = Vec::new();
	for lang in LANGUAGES {
		for pattern in lang.file_patterns {
			let mut glob_builder = GlobBuilder::new(pattern);
			glob_builder.case_insensitive(true);
			let glob = glob_builder
				.build()
				.unwrap_or_else(|e| panic!("Invalid glob pattern '{pattern}' for language {}: {e}", lang.name));
			// Record the language index so glob match indices can map back to LANGUAGES.
			pattern_lang_indexes.push(lang.index);
			builder.add(glob);
		}
	}
	let set = builder.build().expect("GlobSetBuilder should never fail with valid patterns from languages.json5");
	LanguageGlobs { set, pattern_lang_indexes }
});

// Thread-local buffer for tracking which languages have been seen during candidate gathering.
// Using thread-local storage avoids allocating a new Vec for every file processed, while remaining safe for parallel execution since each thread gets its own buffer.
thread_local! {
	static SEEN_BUFFER: RefCell<Vec<bool>> = RefCell::new(vec![false; LANGUAGES.len()]);
}

#[inline]
pub fn get_candidates(filename: &str) -> Vec<&'static Language> {
	let globs = &*LANGUAGE_GLOBSET;
	SEEN_BUFFER.with(|seen_cell| {
		let mut seen = seen_cell.borrow_mut();
		seen.fill(false);
		let mut candidates = Vec::new();
		for match_idx in globs.set.matches(filename) {
			let lang_idx = globs.pattern_lang_indexes[match_idx];
			if !seen[lang_idx] {
				seen[lang_idx] = true;
				candidates.push(&LANGUAGES[lang_idx]);
			}
		}
		candidates
	})
}