Fix memory leak in trie.

6 files changed, 10 insertions, 64 deletions

diff --git a/extractor/matchings_trie.cc b/extractor/matchings_trie.cc
index 921ec582..8ea795db 100644
--- a/extractor/matchings_trie.cc
+++ b/extractor/matchings_trie.cc
@@ -14,6 +14,9 @@ void MatchingsTrie::ResetTree(shared_ptr<TrieNode> root) {
     for (auto child: root->children) {
       ResetTree(child.second);
     }
+    if (root->suffix_link != NULL) {
+      root->suffix_link.reset();
+    }
     root.reset();
   }
 }
diff --git a/extractor/mocks/mock_precomputation.h b/extractor/mocks/mock_precomputation.h
index 987bdb2f..9bc72235 100644
--- a/extractor/mocks/mock_precomputation.h
+++ b/extractor/mocks/mock_precomputation.h
@@ -4,6 +4,5 @@
 
 class MockPrecomputation : public Precomputation {
  public:
-  MOCK_CONST_METHOD0(GetInvertedIndex, const Index&());
   MOCK_CONST_METHOD0(GetCollocations, const Index&());
 };
diff --git a/extractor/precomputation.cc b/extractor/precomputation.cc
index 8a76beb1..189ac42c 100644
--- a/extractor/precomputation.cc
+++ b/extractor/precomputation.cc
@@ -41,7 +41,6 @@ Precomputation::Precomputation(
     for (int j = 1; j <= max_frequent_phrase_len && i + j <= data.size(); ++j) {
       pattern.push_back(data[i + j - 1]);
       if (frequent_patterns_set.count(pattern)) {
-        inverted_index[pattern].push_back(i);
         int is_super_frequent = super_frequent_patterns_set.count(pattern);
         matchings.push_back(make_tuple(i, j, is_super_frequent));
       } else {
@@ -158,19 +157,7 @@ void Precomputation::WriteBinary(const fs::path& filepath) const {
   FILE* file = fopen(filepath.string().c_str(), "w");
 
   // TODO(pauldb): Refactor this code.
-  int size = inverted_index.size();
-  fwrite(&size, sizeof(int), 1, file);
-  for (auto entry: inverted_index) {
-    size = entry.first.size();
-    fwrite(&size, sizeof(int), 1, file);
-    fwrite(entry.first.data(), sizeof(int), size, file);
-
-    size = entry.second.size();
-    fwrite(&size, sizeof(int), 1, file);
-    fwrite(entry.second.data(), sizeof(int), size, file);
-  }
-
-  size = collocations.size();
+  int size = collocations.size();
   fwrite(&size, sizeof(int), 1, file);
   for (auto entry: collocations) {
     size = entry.first.size();
@@ -183,10 +170,6 @@ void Precomputation::WriteBinary(const fs::path& filepath) const {
   }
 }
 
-const Index& Precomputation::GetInvertedIndex() const {
-  return inverted_index;
-}
-
 const Index& Precomputation::GetCollocations() const {
   return collocations;
 }
diff --git a/extractor/precomputation.h b/extractor/precomputation.h
index 28426bfa..3d44c2a6 100644
--- a/extractor/precomputation.h
+++ b/extractor/precomputation.h
@@ -30,7 +30,6 @@ class Precomputation {
 
   void WriteBinary(const fs::path& filepath) const;
 
-  virtual const Index& GetInvertedIndex() const;
   virtual const Index& GetCollocations() const;
 
   static int NON_TERMINAL;
@@ -49,7 +48,6 @@ class Precomputation {
   void AddStartPositions(vector<int>& positions, int pos1, int pos2);
   void AddStartPositions(vector<int>& positions, int pos1, int pos2, int pos3);
 
-  Index inverted_index;
   Index collocations;
 };
 
diff --git a/extractor/precomputation_test.cc b/extractor/precomputation_test.cc
index 9edb29db..6b77b9c0 100644
--- a/extractor/precomputation_test.cc
+++ b/extractor/precomputation_test.cc
@@ -35,40 +35,6 @@ class PrecomputationTest : public Test {
   shared_ptr<MockSuffixArray> suffix_array;
 };
 
-TEST_F(PrecomputationTest, TestInvertedIndex) {
-  Precomputation precomputation(suffix_array, 100, 3, 10, 5, 1, 4, 2);
-  Index inverted_index = precomputation.GetInvertedIndex();
-
-  EXPECT_EQ(8, inverted_index.size());
-  vector<int> key = {2};
-  vector<int> expected_value = {1, 5, 8, 11};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {3};
-  expected_value = {2, 6, 9};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {4};
-  expected_value = {0, 10};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {5};
-  expected_value = {3, 7};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {4, 2};
-  expected_value = {0, 10};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {2, 3};
-  expected_value = {1, 5, 8};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {3, 5};
-  expected_value = {2, 6};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-  key = {2, 3, 5};
-  expected_value = {1, 5};
-  EXPECT_EQ(expected_value, inverted_index[key]);
-
-  key = {2, 4};
-  EXPECT_EQ(0, inverted_index.count(key));
-}
-
 TEST_F(PrecomputationTest, TestCollocations) {
   Precomputation precomputation(suffix_array, 3, 3, 10, 5, 1, 4, 2);
   Index collocations = precomputation.GetCollocations();
diff --git a/extractor/run_extractor.cc b/extractor/run_extractor.cc
index 36dbd7a0..5255737d 100644
--- a/extractor/run_extractor.cc
+++ b/extractor/run_extractor.cc
@@ -31,14 +31,6 @@ namespace fs = boost::filesystem;
 namespace po = boost::program_options;
 using namespace std;
 
-void my_pause() {
-  cerr << "pausing..." << endl;
-  for (int i = 0; i < 10000000; ++i) {
-    cerr << endl;
-  }
-  cerr << "end pause" << endl;
-}
-
 int main(int argc, char** argv) {
   // TODO(pauldb): Also take arguments from config file.
   po::options_description desc("Command line options");
@@ -122,7 +114,7 @@ int main(int argc, char** argv) {
   cerr << "Reading alignment took "
        << GetDuration(start_time, stop_time) << " seconds" << endl;
 
-  cerr << "Precomputating collocations..." << endl;
+  cerr << "Precomputing collocations..." << endl;
   start_time = Clock::now();
   shared_ptr<Precomputation> precomputation = make_shared<Precomputation>(
       source_suffix_array,
@@ -150,6 +142,8 @@ int main(int argc, char** argv) {
        << GetDuration(preprocess_start_time, preprocess_stop_time)
        << " seconds" << endl;
 
+  cerr << "creating grammar extractor" << endl;
+
   Clock::time_point extraction_start_time = Clock::now();
   vector<shared_ptr<Feature> > features = {
       make_shared<TargetGivenSourceCoherent>(),
@@ -176,6 +170,9 @@ int main(int argc, char** argv) {
       vm["max_samples"].as<int>(),
       vm["tight_phrases"].as<bool>());
 
+  // Release extra memory used by the initial precomputation.
+  precomputation.reset();
+
   int grammar_id = 0;
   fs::path grammar_path = vm["grammars"].as<string>();
   if (!fs::is_directory(grammar_path)) {