fix(score): size chunk ptr buffers to PATH_MAX instead of 512 bytes

Cargo.toml

@@ -36,7 +36,7 @@ signal-hook-registry = "1.4"
 zlob = { version = "=1.6.0-dev.7" }
 
 mlua = { version = "0.11.1", features = ["module", "luajit"] }
-neo_frizbee = { version = "0.10.3", features = ["match_end_col"] }
+neo_frizbee = { version = "0.10.4", features = ["match_end_col"] }
 notify = { version = "9.0.0-rc.3" }
 notify-debouncer-full = { package = "fff-notify-debouncer-full", version = "0.9.4" }
 once_cell = "1.20.2"

crates/fff-core/src/score.rs

@@ -2,7 +2,7 @@ use crate::{
     constraints::apply_constraints,
     git::is_modified_status,
     path_utils::calculate_distance_penalty,
-    simd_path::ArenaPtr,
+    simd_path::{ArenaPtr, MAX_PATH_CHUNKS},
     sort_buffer::{sort_by_key_with_buffer, sort_with_buffer},
     types::{DirItem, FileItem, Score, ScoringContext},
 };
@@ -32,7 +32,7 @@ impl<'a> FileItems<'a> {
 fn resolve_file_chunks(
     file: &FileItem,
     arena: ArenaPtr,
-    buf: &mut [*const u8; 32],
+    buf: &mut [*const u8; MAX_PATH_CHUNKS],
 ) -> Option<(usize, u16)> {
     if file.is_deleted() {
         return None;
@@ -60,15 +60,15 @@ fn match_fuzzy_parts(
         return vec![];
     }
 
-    let resolve = |file: &FileItem, buf: &mut [*const u8; 32]| -> Option<(usize, u16)> {
-        resolve_file_chunks(file, arena, buf)
-    };
+    let resolve = |file: &FileItem,
+                   buf: &mut [*const u8; MAX_PATH_CHUNKS]|
+     -> Option<(usize, u16)> { resolve_file_chunks(file, arena, buf) };
 
     // because we reassemble the vec of reference we have to use a different type
     // to narrow down the [&FileItem] which would be resolved by frizbee as &&
-    let resolve_ref = |file: &&FileItem, buf: &mut [*const u8; 32]| -> Option<(usize, u16)> {
-        resolve_file_chunks(file, arena, buf)
-    };
+    let resolve_ref = |file: &&FileItem,
+                       buf: &mut [*const u8; MAX_PATH_CHUNKS]|
+     -> Option<(usize, u16)> { resolve_file_chunks(file, arena, buf) };
 
     let first_part_matches = match working_files {
         FileItems::All(files) => neo_frizbee::match_list_parallel_resolved(
@@ -173,7 +173,7 @@ pub(crate) fn fuzzy_match_and_score_files<'a>(
 fn resolve_dir_chunks(
     dir: &DirItem,
     arena: ArenaPtr,
-    buf: &mut [*const u8; 32],
+    buf: &mut [*const u8; MAX_PATH_CHUNKS],
 ) -> Option<(usize, u16)> {
     let ptrs = dir.path.resolve_ptrs(arena, buf);
     Some((ptrs.len(), dir.path.byte_len))
@@ -199,7 +199,7 @@ fn match_fuzzy_parts_dirs(
     }
 
     let resolve_chunks_for_frizbee =
-        |dir: &&DirItem, buf: &mut [*const u8; 32]| -> Option<(usize, u16)> {
+        |dir: &&DirItem, buf: &mut [*const u8; MAX_PATH_CHUNKS]| -> Option<(usize, u16)> {
             resolve_dir_chunks(dir, arena, buf)
         };
 
@@ -1315,6 +1315,31 @@ mod filename_bonus_tests {
         );
     }
 
+    /// Regression: PR #652 / field panic in pi-fff v0.9.6.
+    /// A path >512 bytes (but within PATH_MAX) overflows the fixed
+    /// `[*const u8; 32]` chunk-pointer buffer during scoring and panics with
+    /// "index out of bounds: the len is 32 but the index is 32".
+    #[test]
+    fn test_path_longer_than_512_bytes_does_not_panic_and_matches() {
+        let mut long_path = String::new();
+        while long_path.len() < 600 {
+            long_path.push_str("deeply_nested_directory_segment/");
+        }
+        long_path.push_str("needle_file.rs");
+        assert!(long_path.len() > 512 && long_path.len() < crate::simd_path::PATH_BUF_SIZE);
+
+        let (files, arena) = make_files(&[long_path.as_str(), "src/other.rs"]);
+
+        // Panics here on unfixed code: frizbee resolves chunk ptrs per file.
+        let results = search(&files, "needle", arena);
+
+        assert!(
+            results.iter().any(|(p, _)| p == &long_path),
+            "filename at the tail of a >512-byte path must still match, got: {:?}",
+            results.iter().map(|(p, _)| p).collect::<Vec<_>>()
+        );
+    }
+
     #[test]
     fn test_single_path_matching() {
         let path = "core_workflow_service/kafka_event_consumer/src/ai_part_extraction_request/ai_part_extraction_request_handler.rs";

crates/fff-core/src/simd_path.rs

@@ -62,6 +62,9 @@ impl std::fmt::Debug for SimdChunk {
 
 pub use crate::constants::PATH_BUF_SIZE;
 
+/// Chunk pointer capacity needed for the longest path the platform allows.
+pub(crate) const MAX_PATH_CHUNKS: usize = PATH_BUF_SIZE.div_ceil(SIMD_CHUNK_BYTES);
+
 /// Indices into a shared `SimdChunk` arena representing a file path.
 ///
 /// All read methods require an explicit `arena_base` pointer from the owning
@@ -98,14 +101,10 @@ impl ChunkedString {
     }
 
     #[inline]
-    pub fn resolve_ptrs<'a>(
-        &self,
-        arena: ArenaPtr,
-        buf: &'a mut [*const u8; 32],
-    ) -> &'a [*const u8] {
-        let count = self.indices.len();
+    pub fn resolve_ptrs<'a>(&self, arena: ArenaPtr, buf: &'a mut [*const u8]) -> &'a [*const u8] {
+        let count = self.indices.len().min(buf.len());
         let base = arena.as_ptr();
-        for (i, &idx) in self.indices.iter().enumerate() {
+        for (i, &idx) in self.indices[..count].iter().enumerate() {
             buf[i] = unsafe { base.add(idx as usize * SIMD_CHUNK_BYTES) };
         }
         &buf[..count]
@@ -460,7 +459,7 @@ mod tests {
         let arena = store.as_arena_ptr();
         let cs = &strings[0];
 
-        let mut ptrs = [std::ptr::null::<u8>(); 32];
+        let mut ptrs = [std::ptr::null::<u8>(); MAX_PATH_CHUNKS];
         let resolved = cs.resolve_ptrs(arena, &mut ptrs);
         assert_eq!(resolved.len(), 2); // 25 bytes = 2 chunks
 
@@ -478,6 +477,43 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_resolve_ptrs_path_exceeding_512_bytes() {
+        // Regression: a fixed 32-ptr buffer covered only 512 bytes while
+        // PATH_BUF_SIZE (libc::PATH_MAX) allows longer paths, panicking with
+        // "index out of bounds: the len is 32 but the index is 32"
+        let mut path = String::new();
+        while path.len() < 600 {
+            path.push_str("deeply_nested_directory_segment/");
+        }
+        path.push_str("needle_file.rs");
+        assert!(path.len() > 512 && path.len() < PATH_BUF_SIZE);
+
+        let (store, strings, _files) = build_test_store(&[path.as_str()]);
+        let arena = store.as_arena_ptr();
+        let cs = &strings[0];
+        assert!(cs.chunk_count() > 32, "path must span more than 32 chunks");
+
+        let mut ptrs = [std::ptr::null::<u8>(); MAX_PATH_CHUNKS];
+        let resolved = cs.resolve_ptrs(arena, &mut ptrs);
+
+        // Truncation is not acceptable either: it silently drops the tail of
+        // the path (including the filename here) from fuzzy matching.
+        assert_eq!(
+            resolved.len(),
+            cs.chunk_count(),
+            "resolve_ptrs must resolve every chunk of a PATH_MAX-legal path"
+        );
+
+        let total = cs.byte_len as usize;
+        let mut reconstructed = Vec::with_capacity(total);
+        for (i, &ptr) in resolved.iter().enumerate() {
+            let take = SIMD_CHUNK_BYTES.min(total - i * SIMD_CHUNK_BYTES);
+            reconstructed.extend_from_slice(unsafe { std::slice::from_raw_parts(ptr, take) });
+        }
+        assert_eq!(std::str::from_utf8(&reconstructed).unwrap(), path);
+    }
+
     #[test]
     fn test_filename_cow_mid_chunk() {
         let (store, strings, _files) = build_test_store(&["src/components/Button.tsx"]);