StrIdx 0.1.4 → 0.1.6

data/stridx.hpp

@@ -10,6 +10,7 @@
 #include <array>
 #include <iostream>
 #include <unordered_map>
+#include <map>
 #include <set>
 #include <algorithm>
 #include <sstream>
@@ -28,10 +29,11 @@ namespace StrIdx {
 class Output {
 private:
   int verboseLevel;
+  // TODO: add mutex?
 
 public:
   Output(int verb) : verboseLevel(verb) {}
-  Output() : Output(3) {}
+  Output() : Output(1) {}
   ~Output() = default;
   static void print() {}
 
@@ -61,61 +63,234 @@ public:
   }
 };
 
+Output out{1};
+
+struct CharNode {
+  int *ids;
+  int ids_sz;
+  char c;
+  std::uint8_t size;
+  CharNode *children;
+  CharNode() : ids(nullptr), ids_sz(0), c(0), size(0), children(nullptr) {}
+
+  void init() {
+    ids = nullptr;
+    ids_sz = 0;
+    c = 0;
+    size = 0;
+    children = nullptr;
+  }
+
+  void dealloc() {
+    if (children != nullptr) {
+      for (CharNode *it = children; it != children + size; it++) {
+        it->dealloc();
+      }
+      free(children);
+    }
+    delete[] ids;
+  }
+
+  ~CharNode() {}
+
+  // Gets Id's stored in this node and all child nodes combined
+  std::set<int> getIds() {
+    std::set<int> set;
+    getIds(set);
+    return set;
+  }
+
+  void getIds(std::set<int> &set) {
+    for (int j = 0; j < ids_sz; j++) {
+      set.insert(ids[j]);
+    }
+    for (CharNode *it = children; it != children + size; it++) {
+      it->getIds(set);
+    }
+  }
+
+  // Find if character 'c' is included in children of the node
+  CharNode *find(char c) {
+    CharNode *ret = nullptr;
+    if (size > 0) {
+      for (auto it = children; it != children + size; it++) {
+        if (it->c == c) {
+          ret = it;
+          break;
+        }
+      }
+    }
+    return ret;
+  }
+};
+
+/* Tree type data structure consisting of strings of file path segments
+ *  (somewhat like a trie)
+ * For example, Adding one input string "abracadabr4" will add the following (size 2..8 char)
+ * substrings: abracada bracadab racadabr acadabr4 dabr4 abr4 br4 ra
+ * (CharTree::addStr callled for each separately)
+ *
+ * Which forms a tree like structure:
+ * [root]-a-b-r-a-c-a-d-a
+ *    |   |   ╰-4
+ *    |   ╰─c-a-d-a-b-r-4
+ *    ╰───b-r-a-c-a-d-a-b
+ *    |     ╰─4
+ *    ╰───r-a-c-a-d-a-b-r
+ *    ╰───d-a-b-r-4
+ *
+ * Id's pointing to path segments are stored in nodes that match the end of the inserted substring
+ *
+ * This data structure (CharTree/CharNode) is the main bottleneck in terms of memory consumption.
+ * For a dataset of 84k files with 3.5 million characters there will be about 2.3 million CharNodes.
+ * Therefore, having std::vector's or similar structures with memory overhead is not really an
+ * option.
+ */
+class CharTree {
+  Output out;
+  std::mutex mu;
+
+public:
+  CharNode *root;
+
+  CharTree() { root = new CharNode; }
+  ~CharTree() {
+    root->dealloc();
+    delete root;
+  }
+
+  void addStr(std::string s, int id) {
+    if (s.size() < 2) {
+      return;
+    }
+
+    // out.printl("add str:",s);
+    CharNode *cn = root;
+
+    std::lock_guard<std::mutex> mu_lock(mu);
+
+    for (int i = 0; i < s.size() && i < 8; i++) {
+      int c = ((char)s[i]);
+      bool found = false;
+
+      if (cn->size > 0) {
+        // out.printl("(1) cn->size > 0");
+        for (auto it = cn->children; it != cn->children + cn->size; it++) {
+          if (it->c == c) {
+            // out.printl("{", c, "}");
+            found = true;
+            cn = it;
+            break;
+          }
+        }
+      }
+      if (!found) {
+        // auto x = new CharNode[cn->size + 1];
+        CharNode *x = (CharNode *)malloc(sizeof(CharNode) * (cn->size + 1));
+        if (cn->size > 0) {
+          memcpy(x, cn->children, sizeof(CharNode) * (cn->size));
+          free(cn->children);
+        }
+        cn->children = x;
+        CharNode *nn = &(cn->children[cn->size]);
+        nn->init();
+        nn->c = c;
+        cn->size++;
+        cn = nn;
+      }
+
+      if (i == s.size() - 1 && true) {
+        out.printv(4, "i=", i, "s:", s.size(), "|");
+        bool found = false;
+        if (cn->ids_sz > 0) {
+          for (int i = 0; i < cn->ids_sz; i++) {
+            if (cn->ids[i] == id) {
+              found = true;
+              out.printv(3, "found:", id, "\n");
+            }
+          }
+        }
+        if (!found) {
+          // out.print(".a.");
+          auto x = new int[cn->ids_sz + 1];
+          if (cn->ids_sz > 0) {
+            memcpy(x, cn->ids, sizeof(int) * cn->ids_sz);
+            delete[] cn->ids;
+          }
+          cn->ids = x;
+          cn->ids[cn->ids_sz] = id;
+          cn->ids_sz++;
+          out.printv(3, "sz:", cn->ids_sz, ",");
+        }
+      }
+
+    } // END for
+  }
+
+  void debug() { debug("", root); }
+  void debug(std::string trail, CharNode *cn) {
+
+    // if (trail.size() > 6) {
+    // out.print("\n");
+    // return;
+    // }
+
+    if (cn == nullptr) {
+      return;
+    }
+    for (int i = 0; i < cn->size; i++) {
+      CharNode *child = &cn->children[i];
+      out.print("[", child->ids_sz, "]");
+      if (child->size > 0) {
+        debug(trail + child->c, child);
+      } else {
+        out.printl(trail, child->c);
+        // out.printl();
+      }
+    }
+  }
+};
+
 // Transforms input string as follows:
 // '/foo/bar/file1.txt'
-// => vector{"foo", "bar", "file1.txt"}
-std::vector<std::string> splitString(const std::string &input, const char &separator) {
+// => vector{"/foo", "/bar", "/file1.txt"}
+
+std::vector<std::string> splitString(const std::string &str, char delimiter) {
   std::vector<std::string> result;
-  std::stringstream ss(input);
-  std::string item;
+  std::string part;
 
-  while (std::getline(ss, item, separator)) {
-    if (item.size() > 0) {
-      result.push_back(item);
+  for (char ch : str) {
+    if (ch == delimiter) {
+      if (part.size() > 0) {
+        result.push_back(part);
+      }
+      part.clear(); // Start a new part
+      part += ch;
+    } else {
+      part += ch;
     }
   }
 
-  return result;
-}
-
-// Convert int64_t to binary string
-[[nodiscard]] std::string int64ToBinaryString(const int64_t &num) {
-  std::string result;
-  for (int i = 63; i >= 0; --i) {
-    result += ((num >> i) & 1) ? '1' : '0';
+  // If there's any remaining part after the loop, add it to the result
+  if (!part.empty()) {
+    result.push_back(part);
   }
-  return result;
-}
 
-// Debug. Convert a (8 char) string represented as int64_t to std::string
-[[nodiscard]] std::string int64ToStr(const int64_t &key) {
-  int nchars = 8;
-  std::string str;
-  int multip = nchars * 8;
-  for (int i = 0; i <= nchars; i++) {
-    char c = (key >> multip) & 255;
-    str.push_back(c);
-    multip -= 8;
-  }
-  return str;
+  // for (const auto &value : result) {
+  // std::cout << value << "|";
+  // }
+  // std::cout << std::endl;
+
+  return result;
 }
 
 // Debug
-void printVector(const std::vector<int> &vec) {
+void printVector(const std::vector<float> &vec) {
   for (const auto &value : vec) {
     std::cout << value << " ";
   }
 }
 
-// Debug
-[[nodiscard]] std::string charToBinaryString(const char &chr) {
-  std::string result;
-  for (int i = 7; i >= 0; --i) {
-    result += ((chr >> i) & 1) ? '1' : '0';
-  }
-  return result;
-}
-
 class Candidate;
 enum class segmentType { Dir, File };
 
@@ -128,7 +303,8 @@ struct PathSegment {
   Candidate *cand;
   PathSegment *parent;
   std::mutex mu;
-  ankerl::unordered_dense::map<std::string, PathSegment *> children;
+  std::map<std::string, PathSegment *> children;
+
   segmentType type = segmentType::Dir;
   PathSegment() : parent(nullptr) {}
   PathSegment(std::string _str) : str(_str), parent(nullptr) {}
@@ -137,9 +313,9 @@ struct PathSegment {
   [[nodiscard]] int size() const {
     int sz = str.size();
     PathSegment *cur = parent;
-    // Sum up length of parent segments (+1 for divisors)
+    // Sum up length of parent segments
     while (cur->parent != nullptr) {
-      sz += cur->str.size() + 1;
+      sz += cur->str.size();
       cur = cur->parent;
     }
     return sz;
@@ -148,7 +324,11 @@ struct PathSegment {
 
 // Candidate for result in string (filename) search
 struct Candidate {
+
+	//This holds the subscores for each character in the query string
   std::vector<float> v_charscore;
+  
+  
   PathSegment *seg;
   int fileId;
   // The string that this candidate represents
@@ -159,6 +339,7 @@ struct Candidate {
   float maxscore;
   int candLen; // Length of candidate
 
+  ~Candidate(){};
   Candidate(){};
   Candidate(PathSegment *_seg, int _len) : seg(_seg), len(_len) {
     // Initialize v_charscores with zeros
@@ -166,6 +347,7 @@ struct Candidate {
     candLen = seg->size();
   }
 
+	// Sum subscores in v_charscore and normalize to get final score
   [[nodiscard]] float getScore() const {
     int i = 0;
     float score = 0.0;
@@ -178,6 +360,7 @@ struct Candidate {
     float div2 = len * candLen;
     float score1 = score / div;
     float score2 = score / div2;
+    // out.printl("str:",seg->str," len:",len," candLen:", candLen, " score:", score);
 
     score = score1 * 0.97 + score2 * 0.03;
     return score;
@@ -186,12 +369,8 @@ struct Candidate {
   [[nodiscard]] float operator[](int idx) const { return v_charscore[idx]; }
 };
 
-// This seems to give 10x speed improvement over std::unordered_map
-typedef ankerl::unordered_dense::map<int64_t, std::set<PathSegment *> *> SegMap;
-// typedef std::unordered_map<int64_t, std::set<PathSegment *> *> SegMap;
-
 typedef ankerl::unordered_dense::map<int, Candidate *> CandMap;
-// typedef std::unordered_map<int, Candidate*> CandMap;
+// typedef std::unordered_map<int, Candidate *> CandMap;
 
 class StringIndex {
 private:
@@ -199,32 +378,29 @@ private:
   char dirSeparator = '/'; // Usually '/', '\' or '\0' (no separator)
   int numStrings = 0;
 
-  std::vector<SegMap *> dirmaps;
-  std::array<std::mutex, 9> mts_d; // for dirmaps
-  std::vector<SegMap *> filemaps;
-  std::array<std::mutex, 9> mts_f; // for filemaps
-
   std::vector<PathSegment *> segsToClean;
 
+	// Maps id's stored in charTree to corresponding PathSegment's
   std::unordered_map<int, PathSegment *> seglist;
+  std::unordered_map<int, PathSegment *> seglist_dir;
+  std::mutex seglist_mu;
+
   PathSegment *root;
   int dirId = 0;
-  float dirWeight = 0.7; // Give only 70% of score if match is for a directory
+  float dirWeight = 1.0; // =0.7: Give only 70% of score if match is for a directory
 
   std::unique_ptr<ThreadPool> pool;
   Output out{1}; // verbose level = 1
+  std::mutex cm_mu;
 
 public:
+  CharTree cm;     // for files
+  CharTree cm_dir; // for directories
   StringIndex(char sep) : dirSeparator(sep) {
     root = new PathSegment();
     root->parent = nullptr;
     root->str = "[ROOT]";
 
-    for (int i = 0; i <= 8; i++) {
-      dirmaps.push_back(new SegMap);
-      filemaps.push_back(new SegMap);
-    }
-
     // Threads between 4 and 6
     // We don't seem to get any benefit from more than 6 threads even if the hardware supports it
     int num_threads = std::max((int)std::thread::hardware_concurrency(), 4);
@@ -233,45 +409,31 @@ public:
     pool = std::unique_ptr<ThreadPool>(new ThreadPool(num_threads));
   }
 
-  /* Don't separate path to segments separator=\0.
+  /* Don't separate path to segments when separator=\0.
      This is slower, but can be used for other data than files also.  */
   StringIndex() : StringIndex('\0') {}
 
   void setDirSeparator(char sep) { dirSeparator = sep; }
   void setDirWeight(float val) { dirWeight = val; }
 
-  ~StringIndex() {
-    for (auto x : dirmaps) {
-      for (auto y : *x) {
-        y.second->clear();
-        delete (y.second);
-      }
-      x->clear();
-      delete x;
-    }
-    for (auto x : filemaps) {
-      for (auto y : *x) {
-        y.second->clear();
-        delete (y.second);
-      }
-      x->clear();
-      delete x;
-    }
-    clearPathSegmentChildren(root);
-  }
+  ~StringIndex() { clearPathSegmentChildren(root); }
 
   void addStrToIndex(std::string filePath, int fileId) {
     addStrToIndex(filePath, fileId, dirSeparator);
   }
 
   void addStrToIndexThreaded(std::string filePath, int fileId) {
-    pool->enqueue([=] { addStrToIndex(filePath, fileId, dirSeparator); });
+    pool->enqueue([filePath, fileId, this] { addStrToIndex(filePath, fileId, dirSeparator); });
+    // addStrToIndex(filePath, fileId, dirSeparator);
   }
   void waitUntilReady() const { pool->waitUntilDone(); }
 
   void waitUntilDone() const { pool->waitUntilDone(); }
 
-  int size() const { return seglist.size(); }
+  int size() {
+    std::lock_guard<std::mutex> guard(seglist_mu);
+    return seglist.size();
+  }
 
   /**
    * Add a string to the index to be searched for afterwards
@@ -283,11 +445,18 @@ public:
    */
 
   void addStrToIndex(std::string filePath, int fileId, const char &separator) {
-    out.printv(3, "Add file:", filePath, ",", fileId, ",", separator, ",",dirSeparator);
+
+    std::lock_guard<std::mutex> guard(cm_mu);
+
+    out.printv(3, "Add file:", filePath, ",", fileId, ",", separator, ",", dirSeparator);
 
     // If a string with this index has beeen added already
-    if (seglist.find(fileId) != seglist.end()) {
-      return;
+    {
+      std::lock_guard<std::mutex> guard(seglist_mu);
+      if (seglist.find(fileId) != seglist.end()) {
+        out.printl("seglist.find(fileId) != seglist.end()");
+        return;
+      }
     }
 
     std::vector<std::string> segs;
@@ -319,36 +488,74 @@ public:
       if (auto it = prev->children.find(x); it != prev->children.end()) {
         p = it->second;
         prev->mu.unlock();
-      } else {
+      } else { // File or dir not included in tree yet
         p = new PathSegment(x, fileId);
         p->parent = prev;
         // If this is last item in segs, then it is a file.
         if (_x == std::prev(segs.end())) {
           p->type = segmentType::File;
-          seglist[fileId] = p;
+          {
+            std::lock_guard<std::mutex> guard(seglist_mu);
+            seglist[fileId] = p;
+
+            for (int i = 0; i < x.size() + 1; i++) {
+              auto s = x.substr(i, std::min(static_cast<size_t>(8), x.size() - i));
+              cm.addStr(s, fileId);
+            }
+          }
         } else { // otherwise, it is a directory
           p->type = segmentType::Dir;
           p->fileId = dirId;
-          // Files use user input Id. Directories need to have it generated
+          /* Add "/" to the end of the string so that
+           * /path/to/file will be indexed as:
+           * {"/path/", "/to/", "/file"}
+           */
+          auto dir_str = x + "/";
+
+          {
+            std::lock_guard<std::mutex> guard(seglist_mu);
+            seglist_dir[dirId] = p;
+            // Files use user input Id. Directories need to have it generated
+          }
+
+          // TODO: Create a function
+          for (int i = 0; i < dir_str.size() + 1; i++) {
+            auto s = dir_str.substr(i, std::min(static_cast<size_t>(8), dir_str.size() - i));
+            cm_dir.addStr(s, dirId);
+          }
+
           dirId++;
         }
         prev->children[x] = p;
         prev->mu.unlock();
-        addPathSegmentKeys(p);
-      }
+      } // END of first if
 
       prev = p;
     }
   }
 
-  std::string getString(int id) {
+  std::string getString(int id) { return getString(id, false); }
+
+  // Reconstruct original filepath from segments
+  std::string getString(int id, bool isDir) {
     std::string s = "";
-    PathSegment *seg = seglist[id];
+    std::lock_guard<std::mutex> guard(seglist_mu);
+
+    PathSegment *seg = nullptr;
+
+    if (isDir) {
+      seg = seglist_dir[id];
+    } else {
+      seg = seglist[id];
+    }
     s += seg->str;
     while (seg->parent->parent != nullptr) {
       seg = seg->parent;
-      s = seg->str + dirSeparator + s;
+      s = seg->str + s;
+      // out.print(seg, "(", seg->str, ")", ",");
     }
+    // out.printl(s);
+
     return s;
   }
 
@@ -384,33 +591,59 @@ public:
   @param query String to search for inside the index
   */
 
-  [[nodiscard]] std::vector<std::pair<float, int>> findSimilar(std::string query) {
-    return findSimilar(query, 2);
-  }
-
-  [[nodiscard]] std::vector<std::pair<float, int>> findSimilar(std::string query, int minChars) {
-    CandMap fileCandMap;
-    CandMap dirCandMap;
-
-    waitUntilDone();
-
-    // Find both files and directories that match the input query
-    addToCandMap(fileCandMap, query, filemaps);
-    addToCandMap(dirCandMap, query, dirmaps);
-
-    /* If parent dir of a file matches the input string add the scores of the direcotry to the
-     scores of the file */
-    mergeCandidateMaps(fileCandMap, dirCandMap);
-
-    // Set all candidate pointers to nullptr so they won't mess up future searches
-    for (auto seg : segsToClean) {
-      seg->cand = nullptr;
+  void searchCharTree(const std::string &query, CandMap &candmap, CharTree &chartr) {
+
+    int last_start = query.size() - 2;
+    // Loop all possible start positions in query string. Indexes [0..(n-3)]
+    for (int start = 0; start <= last_start; start++) {
+      CharNode *cn = chartr.root;
+      
+      // select a suffix (substring) starting from start, but cap length to 8 chars
+      int end = std::min(start + 7, ((int)query.size()) - 1);
+      int nchars = end - start + 1;
+      std::string s = query.substr(start, nchars);
+
+			// Loop all chars of the query substring
+			// Traverse from the 
+      for (int i = 0; i < s.size(); i++) {
+        char c = s[i];
+        CharNode *x = cn->find(c);
+        if (x != nullptr) {
+          cn = x;
+          // Consider scores only for substrings with size >= 2
+          if (i > 0) {
+          	// If we've reached here, size of substring is i+2
+          
+          	// Get identifiers of files that include substring
+          	// query[start..(start+i+1)] ??
+            std::set<int> ids = cn->getIds();
+            for (const int &y : ids) {
+              PathSegment *p = nullptr;
+
+							// Searching in file segments 
+							// (or no file/dir separation)
+              if (&chartr == &cm) {
+                p = seglist[y];
+              } else {
+							// Searching in dir segments 
+                p = seglist_dir[y];
+              }
+              assert(p != nullptr);
+              addToResults(p, query, start, i + 1, candmap);
+            }
+          }
+        } else {
+          // assert(cn->ids_sz < 1); // TODO: should not come here?
+          break;
+        }
+      }
     }
-    segsToClean.clear();
+  }
 
+  std::vector<std::pair<float, int>> candidatesToVec(CandMap &candmap) {
     // Form return result, 2d array with file id's and scores
     std::vector<std::pair<float, int>> results;
-    for (auto &[fid, cand] : fileCandMap) {
+    for (auto &[fid, cand] : candmap) {
       std::pair<float, int> v;
       float sc = cand->getScore();
       v.first = sc;
@@ -419,164 +652,134 @@ public:
       delete cand;
     }
 
-    for (auto &[fid, cand] : dirCandMap) {
-      delete cand;
-    }
-
     // Sort highest score first
     std::sort(results.begin(), results.end(),
               [](std::pair<float, int> a, std::pair<float, int> b) { return a.first > b.first; });
     return results;
   }
 
-  // Return int64_t representation of the first nchars in str, starting from index i
-  [[nodiscard]] int64_t getKeyAtIdx(const std::string &str, int i, int nchars) const {
-    int64_t key = 0;
-    for (int i_char = 0; i_char < nchars; i_char++) {
-      key = key | static_cast<int64_t>(str[i + i_char]);
-      if (i_char < nchars - 1) {
-        // Shift 8 bits to the left except on the last iteration
-        key = key << 8;
-      }
-    }
-    return key;
-  }
+  std::vector<std::pair<float, int>> findDirectories(std::string query) {
+    CandMap dirCandMap;
+    auto &candmap = dirCandMap;
+    waitUntilDone();
 
-  void debug() {
+    searchCharTree(query, dirCandMap, cm_dir);
+    addParentScores(dirCandMap);
+    auto results = candidatesToVec(dirCandMap);
 
-    int nchars = 3;
-    for (const auto &[key, value] : (*filemaps[nchars])) {
-      int64_t x;
-      x = key;
-      int multip = nchars * 8;
-      for (int i = 0; i <= nchars; i++) {
-        char c = (x >> multip) & 255;
-        std::cout << c;
-        multip -= 8;
-      }
-      std::cout << "\n";
-      // for (auto y : *value) {
-      // std::cout << y << " ";
-      // }
-      // std::cout << "\n";
-    }
+    return results;
   }
 
-private:
-  void clearPathSegmentChildren(PathSegment *p) {
-    if (p->children.size() > 0) {
-      for (auto x : p->children) {
-        clearPathSegmentChildren(x.second);
-      }
-    }
-    delete p;
+  std::vector<std::pair<float, std::string>> findFilesAndDirectories(std::string query) {
+    return findFilesAndDirectories(query, true, true);
   }
 
-  void addPathSegmentKeys(PathSegment *p) {
-    // Input p is part of a path, e.g. 'barxyz' if path is /foo/barxyz/baz.txt
-    // This function generates int64 representations (keys) of all substrings of size 2..8 in that
-    // path segment and stores pointer to p in hash tables using these int values as keys.
+  std::vector<std::pair<float, std::string>>
+  findFilesAndDirectories(std::string query, bool includeFiles, bool includeDirs) {
 
-    int maxChars = 8;
-    int minChars = 2;
+    CandMap fileCandMap;
+    CandMap dirCandMap;
+    waitUntilDone();
+    std::vector<std::pair<float, std::string>> results;
 
-    std::string str = p->str;
-    if (p->str.size() < 2) {
-      return;
-    }
-    if (static_cast<int>(p->str.size()) < maxChars) {
-      maxChars = p->str.size();
+    if (includeFiles) {
+      searchCharTree(query, fileCandMap, cm);
+      // out.printl("size:",fileCandMap.size());
     }
 
-    for (int sublen = minChars; sublen <= maxChars; sublen++) {
+    searchCharTree(query, dirCandMap, cm_dir);
 
-      std::mutex *mu;
-      SegMap *map;
-      if (p->type == segmentType::File) {
-        map = filemaps[sublen];
-        mu = &mts_f[sublen];
-      } else {
-        map = dirmaps[sublen];
-        mu = &mts_d[sublen];
-      }
+    if (includeFiles) {
+      addParentScores(fileCandMap);
+    }
 
-      int count = str.size() - sublen + 1;
+    if (includeDirs) {
+      addParentScores(dirCandMap);
+    }
+
+    for (auto seg : segsToClean) {
+      seg->cand = nullptr;
+    }
+    segsToClean.clear();
 
-      int64_t keys[count + 1];
-      for (int i = 0; i <= count; i++) {
-        keys[i] = getKeyAtIdx(str, i, sublen);
+    // TODO: Need to call this just to delete candidates
+    auto res_dir = candidatesToVec(dirCandMap);
+    if (includeDirs) {
+      for (const auto &[score, id] : res_dir) {
+        results.push_back(std::pair<float, std::string>{score, getString(id, true)});
       }
+    }
 
-      mu->lock();
-      for (int i = 0; i <= count; i++) {
-        // int64_t key = getKeyAtIdx(str, i, sublen);
-        auto key = keys[i];
+    if (includeFiles) {
+      auto res_file = candidatesToVec(fileCandMap);
+      // out.printl("size2:",fileCandMap.size());
+      for (const auto &[score, id] : res_file) {
 
-        // Create a new std::set for key if doesn't exist already
-        auto it = map->find(key);
-        if (it == map->end()) {
-          (*map)[key] = new std::set<PathSegment *>;
-        }
-        (*map)[key]->insert(p);
+        // out.print("|",getString(id),"|");
+        results.push_back(std::pair<float, std::string>{score, getString(id)});
       }
-      mu->unlock();
     }
+
+    // Sort highest score first
+    std::sort(results.begin(), results.end(),
+              [](std::pair<float, std::string> a, std::pair<float, std::string> b) {
+                return a.first > b.first;
+              });
+    return results;
   }
 
-  // Find pathsegments from <map> that include the substring of <str> which starts at index <i> and
-  // is of length <nchars>.
-  [[nodiscard]] std::vector<PathSegment *> findSimilarForNgram(std::string str, int i, int nchars,
-                                                               SegMap &map) const {
+  // TODO: delete?
+  std::vector<std::pair<float, int>> findSimilar(std::string query) { return findFiles(query); }
 
-    assert(i + nchars <= static_cast<int>(str.size()));
-    std::vector<PathSegment *> res;
+  std::vector<std::pair<float, int>> findFiles(std::string query) {
 
-    // Take substring of str, starting at i, spanning nchars
-    // transform that to 64 bit integer
-    int64_t key = getKeyAtIdx(str, i, nchars);
-    // Find all path segments in map that have the same substring
-    if (auto it = map.find(key); it != map.end()) { // key found
-      auto set = it->second;
-      for (auto value : *set) {
-        res.push_back(value);
-      }
-    }
-    return res;
-  }
+    CandMap fileCandMap;
+    CandMap dirCandMap;
+    auto &candmap = fileCandMap;
+    waitUntilDone();
+
+    searchCharTree(query, fileCandMap, cm);
+    searchCharTree(query, dirCandMap, cm_dir);
+    addParentScores(fileCandMap);
 
-  void addToCandMap(CandMap &candmap, std::string query,
-                    std::vector<SegMap *> &map // filemaps or dirmaps
-  ) {
-    int maxChars = 8;
-    int minChars = 2;
-    if (static_cast<int>(query.size()) < maxChars) {
-      maxChars = query.size();
+    for (auto seg : segsToClean) {
+      seg->cand = nullptr;
     }
+    segsToClean.clear();
 
-    // Loop all substring lengths between minChars..maxChars
-    for (int sublen = minChars; sublen <= maxChars; sublen++) {
-      int count = query.size() - sublen + 1;
+    auto results = candidatesToVec(fileCandMap);
+    auto tmp = candidatesToVec(dirCandMap); // TODO: call just to release memory
 
-      // Loop all possible start positions
-      for (int i = 0; i < count; i++) {
-        std::vector<PathSegment *> res = findSimilarForNgram(query, i, sublen, *(map[sublen]));
+    return results;
+  }
 
-        for (PathSegment *p : res) {
-          addToResults(p, query, i, sublen, candmap);
-        }
+private:
+  void clearPathSegmentChildren(PathSegment *p) {
+    if (p->children.size() > 0) {
+      for (auto x : p->children) {
+        clearPathSegmentChildren(x.second);
       }
     }
+    delete p;
   }
 
   // Add parent directories scores to files
-  void mergeCandidateMaps(CandMap &fileCandMap, CandMap &dirCandMap) {
+  void addParentScores(CandMap &fileCandMap) {
 
     for (auto &[fid, cand] : fileCandMap) {
       PathSegment *p = cand->seg->parent;
       while (p->parent != nullptr) {
         if (p->cand != nullptr) {
+
           auto &scoreA = cand->v_charscore;
           auto &scoreB = p->cand->v_charscore;
+
+          // out.print("[");
+          // printVector(scoreA);
+          // out.print(",");
+          // printVector(scoreB);
+          // out.print(",");
+          // out.print("]");
           for (int i = 0; i < cand->len; i++) {
             if (scoreA[i] < scoreB[i] * dirWeight) {
               scoreA[i] = scoreB[i] * dirWeight;
@@ -592,6 +795,7 @@ private:
 
     if (auto it2 = candmap.find(seg->fileId); it2 == candmap.end()) {
       Candidate *cand = new Candidate(seg, str.size());
+      // out.printl("new cand:", seg->str, ",", seg, ",", seg->parent, ",", seg->parent->parent);
       segsToClean.push_back(seg);
       candmap[seg->fileId] = cand;
       seg->cand = cand;