1use std::collections::HashMap;
2use std::path::{Path, PathBuf};
3use std::time::UNIX_EPOCH;
4
5use serde::{Deserialize, Serialize};
6
7const MAX_BM25_FILES: usize = 5000;
8const CHUNK_COUNT_WARNING: usize = 50_000;
9const ZSTD_LEVEL: i32 = 9;
10
11const DEFAULT_BM25_IGNORES: &[&str] = &[
12 "vendor/**",
13 "dist/**",
14 "build/**",
15 "public/vendor/**",
16 "public/js/**",
17 "public/css/**",
18 "public/build/**",
19 ".next/**",
20 ".nuxt/**",
21 "__pycache__/**",
22 "*.min.js",
23 "*.min.css",
24 "*.bundle.js",
25 "*.chunk.js",
26];
27
28fn max_bm25_cache_bytes() -> u64 {
29 let mb = std::env::var("LEAN_CTX_BM25_MAX_CACHE_MB")
30 .ok()
31 .and_then(|v| v.parse::<u64>().ok())
32 .unwrap_or_else(|| {
33 let cfg = crate::core::config::Config::load();
34 let profile = crate::core::config::MemoryProfile::effective(&cfg);
35 let profile_mb = profile.bm25_max_cache_mb();
36 if cfg.bm25_max_cache_mb == crate::core::config::default_bm25_max_cache_mb() {
37 profile_mb
38 } else {
39 cfg.bm25_max_cache_mb
40 }
41 });
42 mb * 1024 * 1024
43}
44
45#[derive(Debug, Clone, Serialize, Deserialize)]
46pub struct CodeChunk {
47 pub file_path: String,
48 pub symbol_name: String,
49 pub kind: ChunkKind,
50 pub start_line: usize,
51 pub end_line: usize,
52 pub content: String,
53 #[serde(default)]
54 pub tokens: Vec<String>,
55 pub token_count: usize,
56}
57
58#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
59pub enum ChunkKind {
60 Function,
61 Struct,
62 Impl,
63 Module,
64 Class,
65 Method,
66 Other,
67}
68
69#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
70pub struct IndexedFileState {
71 pub mtime_ms: u64,
72 pub size_bytes: u64,
73}
74
75impl IndexedFileState {
76 fn from_path(path: &Path) -> Option<Self> {
77 let meta = path.metadata().ok()?;
78 let size_bytes = meta.len();
79 let mtime_ms = meta
80 .modified()
81 .ok()
82 .and_then(|t| t.duration_since(UNIX_EPOCH).ok())
83 .map(|d| d.as_millis() as u64)?;
84 Some(Self {
85 mtime_ms,
86 size_bytes,
87 })
88 }
89}
90
91#[derive(Debug, Clone, Serialize, Deserialize)]
92pub struct BM25Index {
93 pub chunks: Vec<CodeChunk>,
94 pub inverted: HashMap<String, Vec<(usize, f64)>>,
95 pub avg_doc_len: f64,
96 pub doc_count: usize,
97 pub doc_freqs: HashMap<String, usize>,
98 #[serde(default)]
99 pub files: HashMap<String, IndexedFileState>,
100}
101
102#[derive(Debug, Clone, Serialize, Deserialize)]
103pub struct SearchResult {
104 pub chunk_idx: usize,
105 pub score: f64,
106 pub file_path: String,
107 pub symbol_name: String,
108 pub kind: ChunkKind,
109 pub start_line: usize,
110 pub end_line: usize,
111 pub snippet: String,
112}
113
114const BM25_K1: f64 = 1.2;
115const BM25_B: f64 = 0.75;
116
117impl Default for BM25Index {
118 fn default() -> Self {
119 Self::new()
120 }
121}
122
123impl BM25Index {
124 pub fn new() -> Self {
125 Self {
126 chunks: Vec::new(),
127 inverted: HashMap::new(),
128 avg_doc_len: 0.0,
129 doc_count: 0,
130 doc_freqs: HashMap::new(),
131 files: HashMap::new(),
132 }
133 }
134
135 pub fn memory_usage_bytes(&self) -> usize {
137 let chunks_size: usize = self
138 .chunks
139 .iter()
140 .map(|c| {
141 c.content.len()
142 + c.file_path.len()
143 + c.symbol_name.len()
144 + c.tokens.iter().map(String::len).sum::<usize>()
145 + 64
146 })
147 .sum();
148 let inverted_size: usize = self
149 .inverted
150 .iter()
151 .map(|(k, v)| k.len() + v.len() * 16 + 32)
152 .sum();
153 let files_size: usize = self.files.keys().map(|k| k.len() + 24).sum();
154 let freqs_size: usize = self.doc_freqs.keys().map(|k| k.len() + 16).sum();
155 chunks_size + inverted_size + files_size + freqs_size
156 }
157
158 pub fn unload(&mut self) {
160 let usage = self.memory_usage_bytes();
161 self.chunks = Vec::new();
162 self.inverted = HashMap::new();
163 self.doc_freqs = HashMap::new();
164 self.files = HashMap::new();
165 self.avg_doc_len = 0.0;
166 self.doc_count = 0;
167 tracing::info!(
168 "[bm25] unloaded index, freed ~{:.1}MB",
169 usage as f64 / 1_048_576.0
170 );
171 }
172
173 #[cfg(test)]
175 pub(crate) fn from_chunks_for_test(chunks: Vec<CodeChunk>) -> Self {
176 let mut index = Self::new();
177 for mut chunk in chunks {
178 if chunk.token_count == 0 {
179 chunk.token_count = tokenize(&chunk.content).len();
180 }
181 index.add_chunk(chunk);
182 }
183 index.finalize();
184 index
185 }
186
187 pub fn build_from_directory(root: &Path) -> Self {
188 let root_str = root.to_string_lossy();
189 if !super::graph_index::is_safe_scan_root_public(&root_str) {
190 tracing::warn!("[bm25: scan aborted for unsafe root {root_str}]");
191 return Self::new();
192 }
193 let mut index = Self::new();
194 let files = list_code_files(root);
195 const MAX_FILE_SIZE_BYTES: u64 = 2 * 1024 * 1024;
196
197 for (i, rel) in files.iter().enumerate() {
198 if i.is_multiple_of(500) && crate::core::memory_guard::is_under_pressure() {
199 tracing::warn!(
200 "[bm25: stopping build at file {i}/{} due to memory pressure]",
201 files.len()
202 );
203 break;
204 }
205 if crate::core::memory_guard::abort_requested() {
206 tracing::warn!("[bm25: aborting build due to critical memory pressure]");
207 break;
208 }
209
210 let abs = root.join(rel);
211 let Some(state) = IndexedFileState::from_path(&abs) else {
212 continue;
213 };
214 if state.size_bytes > MAX_FILE_SIZE_BYTES {
215 continue;
216 }
217 if let Ok(content) = std::fs::read_to_string(&abs) {
218 let mut chunks = extract_chunks(rel, &content);
219 chunks.sort_by(|a, b| {
220 a.start_line
221 .cmp(&b.start_line)
222 .then_with(|| a.end_line.cmp(&b.end_line))
223 .then_with(|| a.symbol_name.cmp(&b.symbol_name))
224 });
225 for chunk in chunks {
226 index.add_chunk(chunk);
227 }
228 index.files.insert(rel.clone(), state);
229 }
230 }
231
232 index.finalize();
233 index
234 }
235
236 pub fn rebuild_incremental(root: &Path, prev: &BM25Index) -> Self {
237 let mut old_by_file: HashMap<String, Vec<CodeChunk>> = HashMap::new();
238 for c in &prev.chunks {
239 old_by_file
240 .entry(c.file_path.clone())
241 .or_default()
242 .push(c.clone());
243 }
244 for v in old_by_file.values_mut() {
245 v.sort_by(|a, b| {
246 a.start_line
247 .cmp(&b.start_line)
248 .then_with(|| a.end_line.cmp(&b.end_line))
249 .then_with(|| a.symbol_name.cmp(&b.symbol_name))
250 });
251 }
252
253 let mut index = Self::new();
254 let files = list_code_files(root);
255 const MAX_FILE_SIZE_BYTES: u64 = 2 * 1024 * 1024;
256
257 for (i, rel) in files.iter().enumerate() {
258 if i.is_multiple_of(500) && crate::core::memory_guard::is_under_pressure() {
259 tracing::warn!(
260 "[bm25: stopping incremental rebuild at file {i}/{} due to memory pressure]",
261 files.len()
262 );
263 break;
264 }
265
266 let abs = root.join(rel);
267 let Some(state) = IndexedFileState::from_path(&abs) else {
268 continue;
269 };
270
271 let unchanged = prev.files.get(rel).is_some_and(|old| *old == state);
272 if unchanged {
273 if let Some(chunks) = old_by_file.get(rel) {
274 if chunks.first().is_some_and(|c| !c.content.is_empty()) {
275 for chunk in chunks {
276 index.add_chunk(chunk.clone());
277 }
278 index.files.insert(rel.clone(), state);
279 continue;
280 }
281 }
282 }
283
284 if state.size_bytes > MAX_FILE_SIZE_BYTES {
285 continue;
286 }
287 if let Ok(content) = std::fs::read_to_string(&abs) {
288 let mut chunks = extract_chunks(rel, &content);
289 chunks.sort_by(|a, b| {
290 a.start_line
291 .cmp(&b.start_line)
292 .then_with(|| a.end_line.cmp(&b.end_line))
293 .then_with(|| a.symbol_name.cmp(&b.symbol_name))
294 });
295 for chunk in chunks {
296 index.add_chunk(chunk);
297 }
298 index.files.insert(rel.clone(), state);
299 }
300 }
301
302 index.finalize();
303 index
304 }
305
306 fn add_chunk(&mut self, chunk: CodeChunk) {
307 let idx = self.chunks.len();
308
309 let enriched = enrich_for_bm25(&chunk);
310 let tokens = tokenize(&enriched);
311 for token in &tokens {
312 let lower = token.to_lowercase();
313 let postings = self.inverted.entry(lower.clone()).or_default();
314 if postings.last().map(|(last_idx, _)| *last_idx) != Some(idx) {
315 *self.doc_freqs.entry(lower).or_insert(0) += 1;
316 }
317 postings.push((idx, 1.0));
318 }
319
320 self.chunks.push(CodeChunk {
321 token_count: tokens.len(),
322 tokens: Vec::new(),
323 ..chunk
324 });
325 }
326
327 fn finalize(&mut self) {
328 self.doc_count = self.chunks.len();
329 if self.doc_count == 0 {
330 return;
331 }
332
333 let total_len: usize = self.chunks.iter().map(|c| c.token_count).sum();
334 self.avg_doc_len = total_len as f64 / self.doc_count as f64;
335 }
336
337 pub fn search(&self, query: &str, top_k: usize) -> Vec<SearchResult> {
338 let query_tokens = tokenize(query);
339 if query_tokens.is_empty() || self.doc_count == 0 {
340 return Vec::new();
341 }
342
343 let n = self.chunks.len();
346 let mut scores = vec![0.0f64; n];
347 let mut touched = Vec::with_capacity(n.min(256));
348
349 for token in &query_tokens {
350 let lower = token.to_lowercase();
351 let df = *self.doc_freqs.get(&lower).unwrap_or(&0) as f64;
352 if df == 0.0 {
353 continue;
354 }
355
356 let idf = ((self.doc_count as f64 - df + 0.5) / (df + 0.5) + 1.0).ln();
357
358 if let Some(postings) = self.inverted.get(&lower) {
359 for &(idx, weight) in postings {
360 let doc_len = self.chunks[idx].token_count as f64;
361 let norm_len = doc_len / self.avg_doc_len.max(1.0);
362 let bm25 = idf * (weight * (BM25_K1 + 1.0))
363 / (weight + BM25_K1 * (1.0 - BM25_B + BM25_B * norm_len));
364
365 if scores[idx] == 0.0 {
366 touched.push(idx);
367 }
368 scores[idx] += bm25;
369 }
370 }
371 }
372
373 let mut results: Vec<SearchResult> = touched
374 .iter()
375 .filter(|&&idx| scores[idx] > 0.0)
376 .map(|&idx| {
377 let chunk = &self.chunks[idx];
378 let snippet = chunk.content.lines().take(5).collect::<Vec<_>>().join("\n");
379 SearchResult {
380 chunk_idx: idx,
381 score: scores[idx],
382 file_path: chunk.file_path.clone(),
383 symbol_name: chunk.symbol_name.clone(),
384 kind: chunk.kind.clone(),
385 start_line: chunk.start_line,
386 end_line: chunk.end_line,
387 snippet,
388 }
389 })
390 .collect();
391
392 results.sort_by(|a, b| {
393 b.score
394 .partial_cmp(&a.score)
395 .unwrap_or(std::cmp::Ordering::Equal)
396 .then_with(|| a.file_path.cmp(&b.file_path))
397 .then_with(|| a.symbol_name.cmp(&b.symbol_name))
398 .then_with(|| a.start_line.cmp(&b.start_line))
399 .then_with(|| a.end_line.cmp(&b.end_line))
400 });
401 results.truncate(top_k);
402 results
403 }
404
405 pub fn save(&self, root: &Path) -> std::io::Result<()> {
406 if self.chunks.len() > CHUNK_COUNT_WARNING {
407 tracing::warn!(
408 "[bm25] index has {} chunks (threshold {}), consider adding extra_ignore_patterns",
409 self.chunks.len(),
410 CHUNK_COUNT_WARNING
411 );
412 }
413
414 let dir = index_dir(root);
415 std::fs::create_dir_all(&dir)?;
416 let data = bincode::serde::encode_to_vec(self, bincode::config::standard())
417 .map_err(|e| std::io::Error::other(e.to_string()))?;
418
419 let compressed = zstd::encode_all(data.as_slice(), ZSTD_LEVEL)
420 .map_err(|e| std::io::Error::other(format!("zstd compress: {e}")))?;
421
422 let max_bytes = max_bm25_cache_bytes();
423 if compressed.len() as u64 > max_bytes {
424 tracing::warn!(
425 "[bm25] compressed index too large ({:.1} MB, limit {:.0} MB), refusing to persist: {}",
426 compressed.len() as f64 / 1_048_576.0,
427 max_bytes / (1024 * 1024),
428 dir.display()
429 );
430 return Ok(());
431 }
432
433 tracing::info!(
434 "[bm25] index: {:.1} MB bincode → {:.1} MB zstd ({:.0}% saved)",
435 data.len() as f64 / 1_048_576.0,
436 compressed.len() as f64 / 1_048_576.0,
437 (1.0 - compressed.len() as f64 / data.len().max(1) as f64) * 100.0
438 );
439
440 let target = dir.join("bm25_index.bin.zst");
441 let tmp = dir.join("bm25_index.bin.zst.tmp");
442 std::fs::write(&tmp, &compressed)?;
443 std::fs::rename(&tmp, &target)?;
444
445 let _ = std::fs::remove_file(dir.join("bm25_index.bin"));
446 let _ = std::fs::remove_file(dir.join("bm25_index.json"));
447
448 let _ = std::fs::write(
449 dir.join("project_root.txt"),
450 root.to_string_lossy().as_bytes(),
451 );
452
453 Ok(())
454 }
455
456 pub fn load(root: &Path) -> Option<Self> {
457 let dir = index_dir(root);
458 let max_bytes = max_bm25_cache_bytes();
459
460 let zst_path = dir.join("bm25_index.bin.zst");
461 if zst_path.exists() {
462 let meta = std::fs::metadata(&zst_path).ok()?;
463 if meta.len() > max_bytes {
464 tracing::warn!(
465 "[bm25] compressed index too large ({:.1} GB, limit {:.0} MB), quarantining: {}",
466 meta.len() as f64 / 1_073_741_824.0,
467 max_bytes / (1024 * 1024),
468 zst_path.display()
469 );
470 let quarantined = zst_path.with_extension("zst.quarantined");
471 let _ = std::fs::rename(&zst_path, &quarantined);
472 return None;
473 }
474 let compressed = std::fs::read(&zst_path).ok()?;
475 let data = zstd::decode_all(compressed.as_slice()).ok()?;
476 let (idx, _): (Self, _) =
477 bincode::serde::decode_from_slice(&data, bincode::config::standard()).ok()?;
478 return Some(idx);
479 }
480
481 let bin_path = dir.join("bm25_index.bin");
482 if bin_path.exists() {
483 let meta = std::fs::metadata(&bin_path).ok()?;
484 if meta.len() > max_bytes {
485 tracing::warn!(
486 "[bm25] index too large ({:.1} GB, limit {:.0} MB), quarantining: {}",
487 meta.len() as f64 / 1_073_741_824.0,
488 max_bytes / (1024 * 1024),
489 bin_path.display()
490 );
491 let quarantined = bin_path.with_extension("bin.quarantined");
492 let _ = std::fs::rename(&bin_path, &quarantined);
493 return None;
494 }
495 let data = std::fs::read(&bin_path).ok()?;
496 let (idx, _): (Self, _) =
497 bincode::serde::decode_from_slice(&data, bincode::config::standard()).ok()?;
498 if let Ok(compressed) = zstd::encode_all(data.as_slice(), ZSTD_LEVEL) {
500 let zst_tmp = zst_path.with_extension("zst.tmp");
501 if std::fs::write(&zst_tmp, &compressed).is_ok()
502 && std::fs::rename(&zst_tmp, &zst_path).is_ok()
503 {
504 tracing::info!(
505 "[bm25] migrated {:.1} MB → {:.1} MB zstd",
506 data.len() as f64 / 1_048_576.0,
507 compressed.len() as f64 / 1_048_576.0
508 );
509 let _ = std::fs::remove_file(&bin_path);
510 }
511 }
512 return Some(idx);
513 }
514
515 let json_path = dir.join("bm25_index.json");
516 if json_path.exists() {
517 let meta = std::fs::metadata(&json_path).ok()?;
518 if meta.len() > max_bytes {
519 tracing::warn!(
520 "[bm25] index too large ({:.1} GB, limit {:.0} MB), quarantining: {}",
521 meta.len() as f64 / 1_073_741_824.0,
522 max_bytes / (1024 * 1024),
523 json_path.display()
524 );
525 let quarantined = json_path.with_extension("json.quarantined");
526 let _ = std::fs::rename(&json_path, &quarantined);
527 return None;
528 }
529 let data = std::fs::read_to_string(&json_path).ok()?;
530 return serde_json::from_str(&data).ok();
531 }
532
533 None
534 }
535
536 pub fn load_or_build(root: &Path) -> Self {
537 if !is_safe_bm25_root(root) {
538 return Self::default();
539 }
540 if let Some(idx) = Self::load(root) {
541 if !bm25_index_looks_stale(&idx, root) {
542 return idx;
543 }
544 tracing::warn!(
545 "[bm25_index: stale index detected for {}; rebuilding]",
546 root.display()
547 );
548 let rebuilt = if idx.files.is_empty() {
549 Self::build_from_directory(root)
550 } else {
551 Self::rebuild_incremental(root, &idx)
552 };
553 let _ = rebuilt.save(root);
554 return rebuilt;
555 }
556
557 let built = Self::build_from_directory(root);
558 let _ = built.save(root);
559 built
560 }
561
562 pub fn index_file_path(root: &Path) -> PathBuf {
563 let dir = index_dir(root);
564 let zst = dir.join("bm25_index.bin.zst");
565 if zst.exists() {
566 return zst;
567 }
568 let bin = dir.join("bm25_index.bin");
569 if bin.exists() {
570 return bin;
571 }
572 dir.join("bm25_index.json")
573 }
574}
575
576fn is_safe_bm25_root(root: &Path) -> bool {
577 super::graph_index::is_safe_scan_root_public(&root.to_string_lossy())
578}
579
580fn bm25_index_looks_stale(index: &BM25Index, root: &Path) -> bool {
581 if index.chunks.is_empty() {
582 return false;
583 }
584
585 if index.files.is_empty() {
586 let mut seen = std::collections::HashSet::<&str>::new();
588 for chunk in &index.chunks {
589 let rel = chunk.file_path.trim_start_matches(['/', '\\']);
590 if rel.is_empty() {
591 continue;
592 }
593 if !seen.insert(rel) {
594 continue;
595 }
596 if !root.join(rel).exists() {
597 return true;
598 }
599 }
600 return false;
601 }
602
603 for (rel, old_state) in &index.files {
605 let abs = root.join(rel);
606 if !abs.exists() {
607 return true;
608 }
609 let Some(cur) = IndexedFileState::from_path(&abs) else {
610 return true;
611 };
612 if &cur != old_state {
613 return true;
614 }
615 }
616
617 for rel in list_code_files(root) {
619 if !index.files.contains_key(&rel) {
620 return true;
621 }
622 }
623
624 false
625}
626
627fn index_dir(root: &Path) -> PathBuf {
628 crate::core::index_namespace::vectors_dir(root)
629}
630
631fn list_code_files(root: &Path) -> Vec<String> {
632 let walker = ignore::WalkBuilder::new(root)
633 .hidden(true)
634 .git_ignore(true)
635 .git_global(true)
636 .git_exclude(true)
637 .max_depth(Some(20))
638 .build();
639
640 let cfg = crate::core::config::Config::load();
641 let mut ignore_patterns: Vec<glob::Pattern> = DEFAULT_BM25_IGNORES
642 .iter()
643 .filter_map(|p| glob::Pattern::new(p).ok())
644 .collect();
645 ignore_patterns.extend(
646 cfg.extra_ignore_patterns
647 .iter()
648 .filter_map(|p| glob::Pattern::new(p).ok()),
649 );
650
651 let mut files: Vec<String> = Vec::new();
652 for entry in walker.flatten() {
653 let path = entry.path();
654 if !path.is_file() {
655 continue;
656 }
657 if !is_code_file(path) {
658 continue;
659 }
660 let rel = path
661 .strip_prefix(root)
662 .unwrap_or(path)
663 .to_string_lossy()
664 .to_string();
665 if rel.is_empty() {
666 continue;
667 }
668 if ignore_patterns.iter().any(|p| p.matches(&rel)) {
669 continue;
670 }
671 if files.len() >= MAX_BM25_FILES {
672 tracing::warn!(
673 "[bm25] file cap reached ({MAX_BM25_FILES}), skipping remaining files in {}",
674 root.display()
675 );
676 break;
677 }
678 files.push(rel);
679 }
680
681 files.sort();
682 files.dedup();
683 files
684}
685
686pub fn is_code_file(path: &Path) -> bool {
687 let ext = path
688 .extension()
689 .and_then(|e| e.to_str())
690 .unwrap_or("")
691 .to_lowercase();
692 matches!(
693 ext.as_str(),
694 "rs" | "ts"
695 | "tsx"
696 | "js"
697 | "jsx"
698 | "py"
699 | "go"
700 | "java"
701 | "c"
702 | "cc"
703 | "cpp"
704 | "h"
705 | "hpp"
706 | "rb"
707 | "cs"
708 | "kt"
709 | "swift"
710 | "php"
711 | "scala"
712 | "sql"
713 | "ex"
714 | "exs"
715 | "zig"
716 | "lua"
717 | "dart"
718 | "vue"
719 | "svelte"
720 )
721}
722
723fn tokenize(text: &str) -> Vec<String> {
724 let mut tokens = Vec::new();
725 let mut current = String::new();
726
727 for ch in text.chars() {
728 if ch.is_alphanumeric() || ch == '_' {
729 current.push(ch);
730 } else {
731 if current.len() >= 2 {
732 tokens.push(current.clone());
733 }
734 current.clear();
735 }
736 }
737 if current.len() >= 2 {
738 tokens.push(current);
739 }
740
741 split_camel_case_tokens(&tokens)
742}
743
744pub(crate) fn tokenize_for_index(text: &str) -> Vec<String> {
745 tokenize(text)
746}
747
748fn split_camel_case_tokens(tokens: &[String]) -> Vec<String> {
749 let mut result = Vec::new();
750 for token in tokens {
751 result.push(token.clone());
752 let mut start = 0;
753 let chars: Vec<char> = token.chars().collect();
754 for i in 1..chars.len() {
755 if chars[i].is_uppercase() && (i + 1 >= chars.len() || !chars[i + 1].is_uppercase()) {
756 let part: String = chars[start..i].iter().collect();
757 if part.len() >= 2 {
758 result.push(part);
759 }
760 start = i;
761 }
762 }
763 if start > 0 {
764 let part: String = chars[start..].iter().collect();
765 if part.len() >= 2 {
766 result.push(part);
767 }
768 }
769 }
770 result
771}
772
773fn extract_chunks(file_path: &str, content: &str) -> Vec<CodeChunk> {
774 #[cfg(feature = "tree-sitter")]
775 {
776 let ext = std::path::Path::new(file_path)
777 .extension()
778 .and_then(|e| e.to_str())
779 .unwrap_or("");
780 if let Some(chunks) = crate::core::chunks_ts::extract_chunks_ts(file_path, content, ext) {
781 return chunks;
782 }
783 }
784
785 let lines: Vec<&str> = content.lines().collect();
786 if lines.is_empty() {
787 return Vec::new();
788 }
789
790 let mut chunks = Vec::new();
791 let mut i = 0;
792
793 while i < lines.len() {
794 let trimmed = lines[i].trim();
795
796 if let Some((name, kind)) = detect_symbol(trimmed) {
797 let start = i;
798 let end = find_block_end(&lines, i);
799 let block: String = lines[start..=end.min(lines.len() - 1)].to_vec().join("\n");
800 let token_count = tokenize(&block).len();
801
802 chunks.push(CodeChunk {
803 file_path: file_path.to_string(),
804 symbol_name: name,
805 kind,
806 start_line: start + 1,
807 end_line: end + 1,
808 content: block,
809 tokens: Vec::new(),
810 token_count,
811 });
812
813 i = end + 1;
814 } else {
815 i += 1;
816 }
817 }
818
819 if chunks.is_empty() && !content.is_empty() {
820 let bytes = content.as_bytes();
825 let rk_chunks = crate::core::rabin_karp::chunk(content);
826 if !rk_chunks.is_empty() && rk_chunks.len() <= 200 {
827 for (idx, c) in rk_chunks.into_iter().take(50).enumerate() {
828 let end = (c.offset + c.length).min(bytes.len());
829 let slice = &bytes[c.offset..end];
830 let chunk_text = String::from_utf8_lossy(slice).into_owned();
831 let token_count = tokenize(&chunk_text).len();
832 let start_line = 1 + bytecount::count(&bytes[..c.offset], b'\n');
833 let end_line = start_line + bytecount::count(slice, b'\n');
834 chunks.push(CodeChunk {
835 file_path: file_path.to_string(),
836 symbol_name: format!("{file_path}#chunk-{idx}"),
837 kind: ChunkKind::Module,
838 start_line,
839 end_line: end_line.max(start_line),
840 content: chunk_text,
841 tokens: Vec::new(),
842 token_count,
843 });
844 }
845 } else {
846 let token_count = tokenize(content).len();
847 let snippet = lines
848 .iter()
849 .take(50)
850 .copied()
851 .collect::<Vec<_>>()
852 .join("\n");
853 chunks.push(CodeChunk {
854 file_path: file_path.to_string(),
855 symbol_name: file_path.to_string(),
856 kind: ChunkKind::Module,
857 start_line: 1,
858 end_line: lines.len(),
859 content: snippet,
860 tokens: Vec::new(),
861 token_count,
862 });
863 }
864 }
865
866 chunks
867}
868
869fn detect_symbol(line: &str) -> Option<(String, ChunkKind)> {
870 let trimmed = line.trim();
871
872 let patterns: &[(&str, ChunkKind)] = &[
873 ("pub async fn ", ChunkKind::Function),
874 ("async fn ", ChunkKind::Function),
875 ("pub fn ", ChunkKind::Function),
876 ("fn ", ChunkKind::Function),
877 ("pub struct ", ChunkKind::Struct),
878 ("struct ", ChunkKind::Struct),
879 ("pub enum ", ChunkKind::Struct),
880 ("enum ", ChunkKind::Struct),
881 ("impl ", ChunkKind::Impl),
882 ("pub trait ", ChunkKind::Struct),
883 ("trait ", ChunkKind::Struct),
884 ("export function ", ChunkKind::Function),
885 ("export async function ", ChunkKind::Function),
886 ("export default function ", ChunkKind::Function),
887 ("function ", ChunkKind::Function),
888 ("async function ", ChunkKind::Function),
889 ("export class ", ChunkKind::Class),
890 ("class ", ChunkKind::Class),
891 ("export interface ", ChunkKind::Struct),
892 ("interface ", ChunkKind::Struct),
893 ("def ", ChunkKind::Function),
894 ("async def ", ChunkKind::Function),
895 ("class ", ChunkKind::Class),
896 ("func ", ChunkKind::Function),
897 ];
898
899 for (prefix, kind) in patterns {
900 if let Some(rest) = trimmed.strip_prefix(prefix) {
901 let name: String = rest
902 .chars()
903 .take_while(|c| c.is_alphanumeric() || *c == '_' || *c == '<')
904 .take_while(|c| *c != '<')
905 .collect();
906 if !name.is_empty() {
907 return Some((name, kind.clone()));
908 }
909 }
910 }
911
912 None
913}
914
915fn find_block_end(lines: &[&str], start: usize) -> usize {
916 let mut depth = 0i32;
917 let mut found_open = false;
918
919 for (i, line) in lines.iter().enumerate().skip(start) {
920 for ch in line.chars() {
921 match ch {
922 '{' | '(' if !found_open || depth > 0 => {
923 depth += 1;
924 found_open = true;
925 }
926 '}' | ')' if depth > 0 => {
927 depth -= 1;
928 if depth == 0 && found_open {
929 return i;
930 }
931 }
932 _ => {}
933 }
934 }
935
936 if found_open && depth <= 0 && i > start {
937 return i;
938 }
939
940 if !found_open && i > start + 2 {
941 let trimmed = lines[i].trim();
942 if trimmed.is_empty()
943 || (!trimmed.starts_with(' ') && !trimmed.starts_with('\t') && i > start)
944 {
945 return i.saturating_sub(1);
946 }
947 }
948 }
949
950 (start + 50).min(lines.len().saturating_sub(1))
951}
952
953pub fn format_search_results(results: &[SearchResult], compact: bool) -> String {
954 if results.is_empty() {
955 return "No results found.".to_string();
956 }
957
958 let mut out = String::new();
959 for (i, r) in results.iter().enumerate() {
960 if compact {
961 out.push_str(&format!(
962 "{}. {:.2} {}:{}-{} {:?} {}\n",
963 i + 1,
964 r.score,
965 r.file_path,
966 r.start_line,
967 r.end_line,
968 r.kind,
969 r.symbol_name,
970 ));
971 } else {
972 out.push_str(&format!(
973 "\n--- Result {} (score: {:.2}) ---\n{} :: {} [{:?}] (L{}-{})\n{}\n",
974 i + 1,
975 r.score,
976 r.file_path,
977 r.symbol_name,
978 r.kind,
979 r.start_line,
980 r.end_line,
981 r.snippet,
982 ));
983 }
984 }
985 out
986}
987
988fn enrich_for_bm25(chunk: &CodeChunk) -> String {
995 let path = Path::new(&chunk.file_path);
996 let stem = path.file_stem().and_then(|s| s.to_str()).unwrap_or("");
997 let dir = path
998 .parent()
999 .and_then(|p| p.file_name())
1000 .and_then(|d| d.to_str())
1001 .unwrap_or("");
1002
1003 if stem.is_empty() {
1004 return chunk.content.clone();
1005 }
1006
1007 format!("{} {} {} {}", chunk.content, stem, stem, dir)
1008}
1009
1010#[cfg(test)]
1011mod tests {
1012 use super::*;
1013 use tempfile::tempdir;
1014
1015 #[cfg(unix)]
1016 use std::os::unix::fs::PermissionsExt;
1017
1018 #[test]
1019 fn tokenize_splits_code() {
1020 let tokens = tokenize("fn calculate_total(items: Vec<Item>) -> f64");
1021 assert!(tokens.contains(&"calculate_total".to_string()));
1022 assert!(tokens.contains(&"items".to_string()));
1023 assert!(tokens.contains(&"Vec".to_string()));
1024 }
1025
1026 #[test]
1027 fn camel_case_splitting() {
1028 let tokens = split_camel_case_tokens(&["calculateTotal".to_string()]);
1029 assert!(tokens.contains(&"calculateTotal".to_string()));
1030 assert!(tokens.contains(&"calculate".to_string()));
1031 assert!(tokens.contains(&"Total".to_string()));
1032 }
1033
1034 #[test]
1035 fn detect_rust_function() {
1036 let (name, kind) =
1037 detect_symbol("pub fn process_request(req: Request) -> Response {").unwrap();
1038 assert_eq!(name, "process_request");
1039 assert_eq!(kind, ChunkKind::Function);
1040 }
1041
1042 #[test]
1043 fn bm25_search_finds_relevant() {
1044 let mut index = BM25Index::new();
1045 index.add_chunk(CodeChunk {
1046 file_path: "auth.rs".into(),
1047 symbol_name: "validate_token".into(),
1048 kind: ChunkKind::Function,
1049 start_line: 1,
1050 end_line: 10,
1051 content: "fn validate_token(token: &str) -> bool { check_jwt_expiry(token) }".into(),
1052 tokens: tokenize("fn validate_token token str bool check_jwt_expiry token"),
1053 token_count: 8,
1054 });
1055 index.add_chunk(CodeChunk {
1056 file_path: "db.rs".into(),
1057 symbol_name: "connect_database".into(),
1058 kind: ChunkKind::Function,
1059 start_line: 1,
1060 end_line: 5,
1061 content: "fn connect_database(url: &str) -> Pool { create_pool(url) }".into(),
1062 tokens: tokenize("fn connect_database url str Pool create_pool url"),
1063 token_count: 7,
1064 });
1065 index.finalize();
1066
1067 let results = index.search("jwt token validation", 5);
1068 assert!(!results.is_empty());
1069 assert_eq!(results[0].symbol_name, "validate_token");
1070 }
1071
1072 #[test]
1073 fn bm25_search_sorts_ties_deterministically() {
1074 let mut index = BM25Index::new();
1075
1076 index.add_chunk(CodeChunk {
1078 file_path: "b.rs".into(),
1079 symbol_name: "same".into(),
1080 kind: ChunkKind::Function,
1081 start_line: 1,
1082 end_line: 1,
1083 content: "fn same() {}".into(),
1084 tokens: tokenize("same token"),
1085 token_count: 2,
1086 });
1087 index.add_chunk(CodeChunk {
1088 file_path: "a.rs".into(),
1089 symbol_name: "same".into(),
1090 kind: ChunkKind::Function,
1091 start_line: 1,
1092 end_line: 1,
1093 content: "fn same() {}".into(),
1094 tokens: tokenize("same token"),
1095 token_count: 2,
1096 });
1097 index.finalize();
1098
1099 let results = index.search("same", 10);
1100 assert!(results.len() >= 2);
1101 assert_eq!(results[0].file_path, "a.rs");
1102 assert_eq!(results[1].file_path, "b.rs");
1103 }
1104
1105 #[test]
1106 fn bm25_index_is_stale_when_any_indexed_file_is_missing() {
1107 let td = tempdir().expect("tempdir");
1108 let root = td.path();
1109 std::fs::write(root.join("a.rs"), "pub fn a() {}\n").expect("write a.rs");
1110
1111 let idx = BM25Index::build_from_directory(root);
1112 assert!(!bm25_index_looks_stale(&idx, root));
1113
1114 std::fs::remove_file(root.join("a.rs")).expect("remove a.rs");
1115 assert!(bm25_index_looks_stale(&idx, root));
1116 }
1117
1118 #[test]
1119 #[cfg(unix)]
1120 fn bm25_incremental_rebuild_reuses_unchanged_files_without_reading() {
1121 let td = tempdir().expect("tempdir");
1122 let root = td.path();
1123
1124 std::fs::write(root.join("a.rs"), "pub fn a() { println!(\"A\"); }\n").expect("write a.rs");
1125 std::fs::write(root.join("b.rs"), "pub fn b() { println!(\"B\"); }\n").expect("write b.rs");
1126
1127 let idx1 = BM25Index::build_from_directory(root);
1128 assert!(idx1.files.contains_key("a.rs"));
1129 assert!(idx1.files.contains_key("b.rs"));
1130
1131 let a_path = root.join("a.rs");
1133 let mut perms = std::fs::metadata(&a_path).expect("meta a.rs").permissions();
1134 perms.set_mode(0o000);
1135 std::fs::set_permissions(&a_path, perms).expect("chmod a.rs");
1136
1137 std::fs::write(root.join("b.rs"), "pub fn b() { println!(\"B2\"); }\n")
1139 .expect("rewrite b.rs");
1140
1141 let idx2 = BM25Index::rebuild_incremental(root, &idx1);
1142 assert!(
1143 idx2.files.contains_key("a.rs"),
1144 "a.rs should be kept via reuse"
1145 );
1146 assert!(idx2.files.contains_key("b.rs"));
1147
1148 let b_has_b2 = idx2
1149 .chunks
1150 .iter()
1151 .any(|c| c.file_path == "b.rs" && c.content.contains("B2"));
1152 assert!(b_has_b2, "b.rs should be re-read and re-chunked");
1153
1154 let mut perms = std::fs::metadata(&a_path).expect("meta a.rs").permissions();
1156 perms.set_mode(0o644);
1157 let _ = std::fs::set_permissions(&a_path, perms);
1158 }
1159
1160 #[test]
1161 fn load_quarantines_oversized_index() {
1162 let _env = crate::core::data_dir::test_env_lock();
1163 let td = tempdir().expect("tempdir");
1164 let root = td.path();
1165 let dir = crate::core::index_namespace::vectors_dir(root);
1166 std::fs::create_dir_all(&dir).expect("create vectors dir");
1167
1168 let index_path = dir.join("bm25_index.json");
1169 std::env::set_var("LEAN_CTX_BM25_MAX_CACHE_MB", "0");
1170 std::fs::write(&index_path, r#"{"chunks":[]}"#).expect("write index");
1171
1172 let result = BM25Index::load(root);
1173 assert!(result.is_none(), "oversized index should return None");
1174 assert!(
1175 !index_path.exists(),
1176 "original index should be removed after quarantine"
1177 );
1178 assert!(
1179 dir.join("bm25_index.json.quarantined").exists(),
1180 "quarantined file should exist"
1181 );
1182
1183 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1184 }
1185
1186 #[test]
1187 fn save_refuses_oversized_output() {
1188 let _env = crate::core::data_dir::test_env_lock();
1189 let data_dir = tempdir().expect("data_dir");
1190 std::env::set_var("LEAN_CTX_DATA_DIR", data_dir.path());
1191 std::env::set_var("LEAN_CTX_BM25_MAX_CACHE_MB", "0");
1192
1193 let td = tempdir().expect("tempdir");
1194 let root = td.path();
1195
1196 let mut index = BM25Index::new();
1197 index.add_chunk(CodeChunk {
1198 file_path: "a.rs".into(),
1199 symbol_name: "a".into(),
1200 kind: ChunkKind::Function,
1201 start_line: 1,
1202 end_line: 1,
1203 content: "fn a() {}".into(),
1204 tokens: tokenize("fn a"),
1205 token_count: 2,
1206 });
1207 index.finalize();
1208
1209 let _ = index.save(root);
1210 let index_path = BM25Index::index_file_path(root);
1211 assert!(
1212 !index_path.exists(),
1213 "save should refuse to persist oversized index"
1214 );
1215
1216 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1217 }
1218
1219 #[test]
1220 fn save_writes_project_root_marker() {
1221 let _env = crate::core::data_dir::test_env_lock();
1222 let td = tempdir().expect("tempdir");
1223 let root = td.path();
1224 std::fs::write(root.join("a.rs"), "pub fn a() {}\n").expect("write");
1225
1226 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1227 let index = BM25Index::build_from_directory(root);
1228 index.save(root).expect("save");
1229
1230 let dir = crate::core::index_namespace::vectors_dir(root);
1231 let marker = dir.join("project_root.txt");
1232 assert!(marker.exists(), "project_root.txt marker should exist");
1233 let content = std::fs::read_to_string(&marker).expect("read marker");
1234 assert_eq!(content, root.to_string_lossy());
1235 }
1236
1237 #[test]
1238 fn save_load_roundtrip_uses_zstd() {
1239 let _env = crate::core::data_dir::test_env_lock();
1240 let data_dir = tempdir().expect("data_dir");
1241 std::env::set_var("LEAN_CTX_DATA_DIR", data_dir.path());
1242 std::env::set_var("LEAN_CTX_BM25_MAX_CACHE_MB", "512");
1243 let td = tempdir().expect("tempdir");
1244 let root = td.path();
1245
1246 for i in 0..10 {
1247 std::fs::write(
1248 root.join(format!("mod{i}.rs")),
1249 format!(
1250 "pub fn handler_{i}() {{\n println!(\"hello\");\n}}\n\n\
1251 pub fn helper_{i}() {{\n println!(\"world\");\n}}\n"
1252 ),
1253 )
1254 .expect("write");
1255 }
1256
1257 let index = BM25Index::build_from_directory(root);
1258 assert!(index.doc_count > 0, "should have indexed chunks");
1259 index.save(root).expect("save");
1260
1261 let dir = crate::core::index_namespace::vectors_dir(root);
1262 let zst = dir.join("bm25_index.bin.zst");
1263 assert!(zst.exists(), "should write .bin.zst");
1264 assert!(
1265 !dir.join("bm25_index.bin").exists(),
1266 ".bin should be deleted"
1267 );
1268
1269 let loaded = BM25Index::load(root).expect("load compressed index");
1270 assert_eq!(loaded.doc_count, index.doc_count);
1271 assert_eq!(loaded.chunks.len(), index.chunks.len());
1272
1273 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1274 std::env::remove_var("LEAN_CTX_DATA_DIR");
1275 }
1276
1277 #[test]
1278 fn auto_migrate_bin_to_zst() {
1279 let _env = crate::core::data_dir::test_env_lock();
1280 let data_dir = tempdir().expect("data_dir");
1281 std::env::set_var("LEAN_CTX_DATA_DIR", data_dir.path());
1282 std::env::set_var("LEAN_CTX_BM25_MAX_CACHE_MB", "512");
1283 let td = tempdir().expect("tempdir");
1284 let root = td.path();
1285
1286 std::fs::write(root.join("a.rs"), "pub fn a() {}\n").expect("write");
1287 let index = BM25Index::build_from_directory(root);
1288
1289 let dir = crate::core::index_namespace::vectors_dir(root);
1290 std::fs::create_dir_all(&dir).expect("mkdir");
1291 let data =
1292 bincode::serde::encode_to_vec(&index, bincode::config::standard()).expect("encode");
1293 std::fs::write(dir.join("bm25_index.bin"), &data).expect("write bin");
1294
1295 let loaded = BM25Index::load(root).expect("load should auto-migrate");
1296 assert_eq!(loaded.doc_count, index.doc_count);
1297 assert!(
1298 dir.join("bm25_index.bin.zst").exists(),
1299 ".bin.zst should be created"
1300 );
1301 assert!(
1302 !dir.join("bm25_index.bin").exists(),
1303 ".bin should be removed"
1304 );
1305
1306 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1307 std::env::remove_var("LEAN_CTX_DATA_DIR");
1308 }
1309
1310 #[test]
1311 fn list_code_files_skips_default_vendor_ignores() {
1312 let td = tempdir().expect("tempdir");
1313 let root = td.path();
1314
1315 std::fs::write(root.join("main.rs"), "pub fn main() {}\n").expect("write main");
1316 std::fs::create_dir_all(root.join("vendor/lib")).expect("mkdir vendor");
1317 std::fs::write(root.join("vendor/lib/dep.rs"), "pub fn dep() {}\n").expect("write vendor");
1318 std::fs::create_dir_all(root.join("dist")).expect("mkdir dist");
1319 std::fs::write(root.join("dist/bundle.js"), "function x() {}").expect("write dist");
1320
1321 let files = list_code_files(root);
1322 assert!(
1323 files.iter().any(|f| f == "main.rs"),
1324 "main.rs should be included"
1325 );
1326 assert!(
1327 !files.iter().any(|f| f.starts_with("vendor/")),
1328 "vendor/ files should be excluded by DEFAULT_BM25_IGNORES"
1329 );
1330 assert!(
1331 !files.iter().any(|f| f.starts_with("dist/")),
1332 "dist/ files should be excluded by DEFAULT_BM25_IGNORES"
1333 );
1334 }
1335
1336 #[test]
1337 fn list_code_files_respects_max_files_cap() {
1338 let td = tempdir().expect("tempdir");
1339 let root = td.path();
1340
1341 for i in 0..10 {
1344 std::fs::write(
1345 root.join(format!("f{i}.rs")),
1346 format!("pub fn f{i}() {{}}\n"),
1347 )
1348 .expect("write");
1349 }
1350 let files = list_code_files(root);
1351 assert!(
1352 files.len() <= MAX_BM25_FILES,
1353 "file count should not exceed MAX_BM25_FILES"
1354 );
1355 }
1356
1357 #[test]
1358 fn max_bm25_cache_bytes_reads_env() {
1359 let _env = crate::core::data_dir::test_env_lock();
1360 std::env::set_var("LEAN_CTX_BM25_MAX_CACHE_MB", "64");
1361 let bytes = max_bm25_cache_bytes();
1362 assert_eq!(bytes, 64 * 1024 * 1024);
1363 std::env::remove_var("LEAN_CTX_BM25_MAX_CACHE_MB");
1364 }
1365}