feat: add fuzzy matching to file search

This commit is contained in:
Bohdan Triapitsyn
2025-12-26 01:18:29 +02:00
parent 2080ce6d65
commit 3c8cc3a93e
3 changed files with 355 additions and 51 deletions
+122 -17
View File
@@ -211,6 +211,11 @@ pub async fn list_directory(
})
}
struct ScoredFileHit {
hit: FileSearchHit,
score: i32,
}
#[tauri::command]
pub async fn search_files(
directory: Option<String>,
@@ -227,14 +232,17 @@ pub async fn search_files(
let normalized_query = query.unwrap_or_default().trim().to_lowercase();
let match_all = normalized_query.is_empty();
let mut files = Vec::new();
// Collect more candidates for fuzzy matching, then sort and trim
let collect_limit = if match_all { limit } else { (limit * 3).max(200) };
let mut candidates: Vec<ScoredFileHit> = Vec::new();
let mut queue = VecDeque::new();
let mut visited = HashSet::new();
queue.push_back(resolved_root.clone());
visited.insert(resolved_root.clone());
while !queue.is_empty() && files.len() < limit {
while !queue.is_empty() && candidates.len() < collect_limit {
for _ in 0..FILE_SEARCH_MAX_CONCURRENCY {
let Some(dir) = queue.pop_front() else {
break;
@@ -261,7 +269,7 @@ pub async fn search_files(
if should_skip_directory(&name_str) {
continue;
}
if visited.insert(entry_path.clone()) && files.len() < limit {
if visited.insert(entry_path.clone()) && candidates.len() < collect_limit {
queue.push_back(entry_path);
}
continue;
@@ -272,39 +280,59 @@ pub async fn search_files(
}
let relative_path = relative_path(&resolved_root, &entry_path);
if !match_all {
let lowercase_name = name_str.to_lowercase();
let lowercase_path = relative_path.to_lowercase();
if !lowercase_name.contains(&normalized_query)
&& !lowercase_path.contains(&normalized_query)
{
continue;
}
}
let extension = entry_path
.extension()
.and_then(|ext| ext.to_str())
.map(|ext| ext.to_lowercase());
files.push(FileSearchHit {
let hit = FileSearchHit {
name: name_str.to_string(),
path: normalize_path(&entry_path),
relative_path: relative_path.replace('\\', "/"),
extension,
});
};
if files.len() >= limit {
if match_all {
candidates.push(ScoredFileHit { hit, score: 0 });
} else {
// Try fuzzy match against relative path (includes filename)
if let Some(score) = fuzzy_match_score(&normalized_query, &relative_path) {
candidates.push(ScoredFileHit { hit, score });
}
}
if candidates.len() >= collect_limit {
break;
}
}
if files.len() >= limit {
if candidates.len() >= collect_limit {
break;
}
}
}
// Sort by score descending, then by path length, then alphabetically
if !match_all {
candidates.sort_by(|a, b| {
match b.score.cmp(&a.score) {
std::cmp::Ordering::Equal => {
match a.hit.relative_path.len().cmp(&b.hit.relative_path.len()) {
std::cmp::Ordering::Equal => a.hit.relative_path.cmp(&b.hit.relative_path),
other => other,
}
}
other => other,
}
});
}
let files: Vec<FileSearchHit> = candidates
.into_iter()
.take(limit)
.map(|scored| scored.hit)
.collect();
Ok(SearchFilesResponse {
root: normalize_path(&resolved_root),
count: files.len(),
@@ -432,6 +460,83 @@ fn should_skip_directory(name: &str) -> bool {
.any(|dir| dir.eq_ignore_ascii_case(name))
}
/// Fuzzy match scoring function.
/// Returns Some(score) if the query fuzzy-matches the candidate, None otherwise.
/// Higher scores indicate better matches.
fn fuzzy_match_score(query: &str, candidate: &str) -> Option<i32> {
if query.is_empty() {
return Some(0);
}
let q: Vec<char> = query.to_lowercase().chars().collect();
let c: Vec<char> = candidate.to_lowercase().chars().collect();
let c_str = candidate.to_lowercase();
// Fast path: exact substring match gets high score
if c_str.contains(query) {
if let Some(idx) = c_str.find(query) {
let mut bonus: i32 = 0;
if idx == 0 {
bonus = 20;
} else if let Some(prev) = c.get(idx.saturating_sub(1)) {
if *prev == '/' || *prev == '_' || *prev == '-' || *prev == '.' || *prev == ' ' {
bonus = 15;
}
}
return Some(100 + bonus - (idx.min(20) as i32) - (c.len() as i32 / 5));
}
}
// Fuzzy match: all query chars must appear in order
let mut score: i32 = 0;
let mut last_index: i32 = -1;
let mut consecutive: i32 = 0;
for ch in &q {
if *ch == ' ' {
continue;
}
let search_start = if last_index < 0 { 0 } else { (last_index + 1) as usize };
let idx = c[search_start..].iter().position(|&c_char| c_char == *ch);
match idx {
None => return None, // No match
Some(relative_idx) => {
let idx = search_start + relative_idx;
let gap = idx as i32 - last_index - 1;
if gap == 0 {
consecutive += 1;
} else {
consecutive = 0;
}
score += 10;
score += (18 - idx as i32).max(0); // Prefer matches near start
score -= gap.min(10); // Penalize gaps
// Bonus for word boundary matches
if idx == 0 {
score += 12;
} else if let Some(prev) = c.get(idx - 1) {
if *prev == '/' || *prev == '_' || *prev == '-' || *prev == '.' || *prev == ' ' {
score += 10;
}
}
score += if consecutive > 0 { 12 } else { 0 }; // Bonus for consecutive matches
last_index = idx as i32;
}
}
}
// Prefer shorter paths
score += (24 - c.len() as i32 / 3).max(0);
Some(score)
}
fn normalize_path(path: &Path) -> String {
path.to_string_lossy().replace('\\', "/")
}
+116 -17
View File
@@ -140,6 +140,77 @@ const shouldSkipSearchDirectory = (name: string) => {
return FILE_SEARCH_EXCLUDED_DIRS.has(name.toLowerCase());
};
/**
* Fuzzy match scoring function.
* Returns a score > 0 if the query fuzzy-matches the candidate, null otherwise.
* Higher scores indicate better matches.
*/
const fuzzyMatchScore = (query: string, candidate: string): number | null => {
if (!query) return 0;
const q = query.toLowerCase();
const c = candidate.toLowerCase();
// Fast path: exact substring match gets high score
if (c.includes(q)) {
const idx = c.indexOf(q);
let bonus = 0;
if (idx === 0) {
bonus = 20;
} else {
const prev = c[idx - 1];
if (prev === '/' || prev === '_' || prev === '-' || prev === '.' || prev === ' ') {
bonus = 15;
}
}
return 100 + bonus - Math.min(idx, 20) - Math.floor(c.length / 5);
}
// Fuzzy match: all query chars must appear in order
let score = 0;
let lastIndex = -1;
let consecutive = 0;
for (let i = 0; i < q.length; i++) {
const ch = q[i];
if (!ch || ch === ' ') continue;
const idx = c.indexOf(ch, lastIndex + 1);
if (idx === -1) {
return null; // No match
}
const gap = idx - lastIndex - 1;
if (gap === 0) {
consecutive++;
} else {
consecutive = 0;
}
score += 10;
score += Math.max(0, 18 - idx); // Prefer matches near start
score -= Math.min(gap, 10); // Penalize gaps
// Bonus for word boundary matches
if (idx === 0) {
score += 12;
} else {
const prev = c[idx - 1];
if (prev === '/' || prev === '_' || prev === '-' || prev === '.' || prev === ' ') {
score += 10;
}
}
score += consecutive > 0 ? 12 : 0; // Bonus for consecutive matches
lastIndex = idx;
}
// Prefer shorter paths
score += Math.max(0, 24 - Math.floor(c.length / 3));
return score;
};
const searchFilesystemFiles = async (rootPath: string, query: string, limit: number) => {
const normalizedQuery = (query || '').trim().toLowerCase();
const matchAll = normalizedQuery.length === 0;
@@ -147,10 +218,12 @@ const searchFilesystemFiles = async (rootPath: string, query: string, limit: num
const rootUri = vscode.Uri.file(rootPath);
const queue: vscode.Uri[] = [rootUri];
const visited = new Set<string>([normalizeFsPath(rootUri.fsPath)]);
const results: Array<{ name: string; path: string; relativePath: string; extension?: string }> = [];
// Collect more candidates for fuzzy matching, then sort and trim
const collectLimit = matchAll ? limit : Math.max(limit * 3, 200);
const candidates: Array<{ name: string; path: string; relativePath: string; extension?: string; score: number }> = [];
const MAX_CONCURRENCY = 5;
while (queue.length > 0 && results.length < limit) {
while (queue.length > 0 && candidates.length < collectLimit) {
const batch = queue.splice(0, MAX_CONCURRENCY);
const dirLists = await Promise.all(
batch.map((dir) => Promise.resolve(vscode.workspace.fs.readDirectory(dir)).catch(() => [] as [string, vscode.FileType][]))
@@ -184,34 +257,60 @@ const searchFilesystemFiles = async (rootPath: string, query: string, limit: num
}
const relativePath = normalizeFsPath(path.relative(rootPath, absolute) || path.basename(absolute));
if (!matchAll) {
const lowercaseName = entryName.toLowerCase();
const lowercasePath = relativePath.toLowerCase();
if (!lowercaseName.includes(normalizedQuery) && !lowercasePath.includes(normalizedQuery)) {
continue;
const extension = entryName.includes('.') ? entryName.split('.').pop()?.toLowerCase() : undefined;
if (matchAll) {
candidates.push({
name: entryName,
path: absolute,
relativePath,
extension,
score: 0,
});
} else {
// Try fuzzy match against relative path (includes filename)
const score = fuzzyMatchScore(normalizedQuery, relativePath);
if (score !== null) {
candidates.push({
name: entryName,
path: absolute,
relativePath,
extension,
score,
});
}
}
results.push({
name: entryName,
path: absolute,
relativePath,
extension: entryName.includes('.') ? entryName.split('.').pop()?.toLowerCase() : undefined,
});
if (results.length >= limit) {
if (candidates.length >= collectLimit) {
queue.length = 0;
break;
}
}
if (results.length >= limit) {
if (candidates.length >= collectLimit) {
break;
}
}
}
return results;
// Sort by score descending, then by path length, then alphabetically
if (!matchAll) {
candidates.sort((a, b) => {
if (b.score !== a.score) return b.score - a.score;
if (a.relativePath.length !== b.relativePath.length) {
return a.relativePath.length - b.relativePath.length;
}
return a.relativePath.localeCompare(b.relativePath);
});
}
// Return top results without the score field
return candidates.slice(0, limit).map(({ name, path: filePath, relativePath, extension }) => ({
name,
path: filePath,
relativePath,
extension,
}));
};
const searchDirectory = async (directory: string, query: string, limit = 60) => {
+117 -17
View File
@@ -81,15 +81,89 @@ const listDirectoryEntries = async (dirPath) => {
}
};
/**
* Fuzzy match scoring function.
* Returns a score > 0 if the query fuzzy-matches the candidate, null otherwise.
* Higher scores indicate better matches.
*/
const fuzzyMatchScore = (query, candidate) => {
if (!query) return 0;
const q = query.toLowerCase();
const c = candidate.toLowerCase();
// Fast path: exact substring match gets high score
if (c.includes(q)) {
const idx = c.indexOf(q);
// Bonus for match at start or after word boundary
let bonus = 0;
if (idx === 0) {
bonus = 20;
} else {
const prev = c[idx - 1];
if (prev === '/' || prev === '_' || prev === '-' || prev === '.' || prev === ' ') {
bonus = 15;
}
}
return 100 + bonus - Math.min(idx, 20) - Math.floor(c.length / 5);
}
// Fuzzy match: all query chars must appear in order
let score = 0;
let lastIndex = -1;
let consecutive = 0;
for (let i = 0; i < q.length; i++) {
const ch = q[i];
if (!ch || ch === ' ') continue;
const idx = c.indexOf(ch, lastIndex + 1);
if (idx === -1) {
return null; // No match
}
const gap = idx - lastIndex - 1;
if (gap === 0) {
consecutive++;
} else {
consecutive = 0;
}
score += 10;
score += Math.max(0, 18 - idx); // Prefer matches near start
score -= Math.min(gap, 10); // Penalize gaps
// Bonus for word boundary matches
if (idx === 0) {
score += 12;
} else {
const prev = c[idx - 1];
if (prev === '/' || prev === '_' || prev === '-' || prev === '.' || prev === ' ') {
score += 10;
}
}
score += consecutive > 0 ? 12 : 0; // Bonus for consecutive matches
lastIndex = idx;
}
// Prefer shorter paths
score += Math.max(0, 24 - Math.floor(c.length / 3));
return score;
};
const searchFilesystemFiles = async (rootPath, options) => {
const { limit, query } = options;
const normalizedQuery = query.trim().toLowerCase();
const matchAll = normalizedQuery.length === 0;
const queue = [rootPath];
const visited = new Set([rootPath]);
const results = [];
// Collect more candidates for fuzzy matching, then sort and trim
const collectLimit = matchAll ? limit : Math.max(limit * 3, 200);
const candidates = [];
while (queue.length > 0 && results.length < limit) {
while (queue.length > 0 && candidates.length < collectLimit) {
const batch = queue.splice(0, FILE_SEARCH_MAX_CONCURRENCY);
const dirLists = await Promise.all(batch.map((dir) => listDirectoryEntries(dir)));
@@ -121,34 +195,60 @@ const searchFilesystemFiles = async (rootPath, options) => {
}
const relativePath = normalizeRelativeSearchPath(rootPath, entryPath);
if (!matchAll) {
const lowercaseName = entryName.toLowerCase();
const lowercasePath = relativePath.toLowerCase();
if (!lowercaseName.includes(normalizedQuery) && !lowercasePath.includes(normalizedQuery)) {
continue;
const extension = entryName.includes('.') ? entryName.split('.').pop()?.toLowerCase() : undefined;
if (matchAll) {
candidates.push({
name: entryName,
path: entryPath,
relativePath,
extension,
score: 0
});
} else {
// Try fuzzy match against relative path (includes filename)
const score = fuzzyMatchScore(normalizedQuery, relativePath);
if (score !== null) {
candidates.push({
name: entryName,
path: entryPath,
relativePath,
extension,
score
});
}
}
results.push({
name: entryName,
path: entryPath,
relativePath,
extension: entryName.includes('.') ? entryName.split('.').pop()?.toLowerCase() : undefined
});
if (results.length >= limit) {
if (candidates.length >= collectLimit) {
queue.length = 0;
break;
}
}
if (results.length >= limit) {
if (candidates.length >= collectLimit) {
break;
}
}
}
return results;
// Sort by score descending, then by path length, then alphabetically
if (!matchAll) {
candidates.sort((a, b) => {
if (b.score !== a.score) return b.score - a.score;
if (a.relativePath.length !== b.relativePath.length) {
return a.relativePath.length - b.relativePath.length;
}
return a.relativePath.localeCompare(b.relativePath);
});
}
// Return top results without the score field
return candidates.slice(0, limit).map(({ name, path: filePath, relativePath, extension }) => ({
name,
path: filePath,
relativePath,
extension
}));
};
const createTimeoutSignal = (timeoutMs) => {