Merge "Allow large memory space for GC."

native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_policy.cpp

@@ -268,8 +268,8 @@ bool DynamicPatriciaTriePolicy::needsToRunGC() const {
         AKLOGI("Warning: needsToRunGC() is called for non-updatable dictionary.");
         return false;
     }
-    // TODO: Implement.
+    // TODO: Implement more properly.
-    return false;
+    return mBufferWithExtendableBuffer.isNearSizeLimit();
 }
 
 } // namespace latinime

native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.cpp

@@ -35,6 +35,8 @@ namespace latinime {
 const int DynamicPatriciaTrieWritingHelper::CHILDREN_POSITION_FIELD_SIZE = 3;
 const char *const DynamicPatriciaTrieWritingHelper::TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE =
         ".tmp";
+// TODO: Make MAX_DICTIONARY_SIZE 8MB.
+const size_t DynamicPatriciaTrieWritingHelper::MAX_DICTIONARY_SIZE = 2 * 1024 * 1024;
 
 bool DynamicPatriciaTrieWritingHelper::addUnigramWord(
         DynamicPatriciaTrieReadingHelper *const readingHelper,
@@ -154,7 +156,8 @@ void DynamicPatriciaTrieWritingHelper::writeToDictFileWithGC(const int rootPtNod
     if (!headerPolicy->writeHeaderToBuffer(&headerBuffer, true /* updatesLastUpdatedTime */)) {
         return;
     }
-    BufferWithExtendableBuffer newDictBuffer(0 /* originalBuffer */, 0 /* originalBufferSize */);
+    BufferWithExtendableBuffer newDictBuffer(0 /* originalBuffer */, 0 /* originalBufferSize */,
+            MAX_DICTIONARY_SIZE);
     if (!runGC(rootPtNodeArrayPos, &newDictBuffer)) {
         return;
     }

native/jni/src/suggest/policyimpl/dictionary/dynamic_patricia_trie_writing_helper.h

@@ -86,6 +86,7 @@ class DynamicPatriciaTrieWritingHelper {
 
     static const int CHILDREN_POSITION_FIELD_SIZE;
     static const char *const TEMP_FILE_SUFFIX_FOR_WRITING_DICT_FILE;
+    static const size_t MAX_DICTIONARY_SIZE;
 
     BufferWithExtendableBuffer *const mBuffer;
     DynamicBigramListPolicy *const mBigramPolicy;

native/jni/src/suggest/policyimpl/dictionary/utils/buffer_with_extendable_buffer.cpp

@@ -18,9 +18,10 @@
 
 namespace latinime {
 
-const size_t BufferWithExtendableBuffer::INITIAL_ADDITIONAL_BUFFER_SIZE = 16 * 1024;
 const size_t BufferWithExtendableBuffer::MAX_ADDITIONAL_BUFFER_SIZE = 1024 * 1024;
-const size_t BufferWithExtendableBuffer::EXTEND_ADDITIONAL_BUFFER_SIZE_STEP = 16 * 1024;
+const int BufferWithExtendableBuffer::NEAR_BUFFER_LIMIT_THRESHOLD_PERCENTILE = 90;
+// TODO: Needs to allocate larger memory corresponding to the current vector size.
+const size_t BufferWithExtendableBuffer::EXTEND_ADDITIONAL_BUFFER_SIZE_STEP = 128 * 1024;
 
 bool BufferWithExtendableBuffer::writeUintAndAdvancePosition(const uint32_t data, const int size,
         int *const pos) {
@@ -64,6 +65,16 @@ bool BufferWithExtendableBuffer::writeCodePointsAndAdvancePosition(const int *co
     return true;
 }
 
+bool BufferWithExtendableBuffer::extendBuffer() {
+    const size_t sizeAfterExtending =
+            mAdditionalBuffer.size() + EXTEND_ADDITIONAL_BUFFER_SIZE_STEP;
+    if (sizeAfterExtending > mMaxAdditionalBufferSize) {
+        return false;
+    }
+    mAdditionalBuffer.resize(mAdditionalBuffer.size() + EXTEND_ADDITIONAL_BUFFER_SIZE_STEP);
+    return true;
+}
+
 bool BufferWithExtendableBuffer::checkAndPrepareWriting(const int pos, const int size) {
     if (isInAdditionalBuffer(pos)) {
         const int tailPosition = getTailPosition();

native/jni/src/suggest/policyimpl/dictionary/utils/buffer_with_extendable_buffer.h

@@ -32,9 +32,11 @@ namespace latinime {
 // raw pointer but provides several methods that handle boundary checking for writing data.
 class BufferWithExtendableBuffer {
  public:
-    BufferWithExtendableBuffer(uint8_t *const originalBuffer, const int originalBufferSize)
+    BufferWithExtendableBuffer(uint8_t *const originalBuffer, const int originalBufferSize,
+            const int maxAdditionalBufferSize = MAX_ADDITIONAL_BUFFER_SIZE)
             : mOriginalBuffer(originalBuffer), mOriginalBufferSize(originalBufferSize),
-              mAdditionalBuffer(INITIAL_ADDITIONAL_BUFFER_SIZE), mUsedAdditionalBufferSize(0) {}
+              mAdditionalBuffer(EXTEND_ADDITIONAL_BUFFER_SIZE_STEP), mUsedAdditionalBufferSize(0),
+              mMaxAdditionalBufferSize(maxAdditionalBufferSize) {}
 
     AK_FORCE_INLINE int getTailPosition() const {
         return mOriginalBufferSize + mUsedAdditionalBufferSize;
@@ -61,6 +63,11 @@ class BufferWithExtendableBuffer {
         return mOriginalBufferSize;
     }
 
+    AK_FORCE_INLINE bool isNearSizeLimit() const {
+        return mAdditionalBuffer.size() >= ((mMaxAdditionalBufferSize
+                * NEAR_BUFFER_LIMIT_THRESHOLD_PERCENTILE) / 100);
+    }
+
     /**
      * For writing.
      *
@@ -75,28 +82,22 @@ class BufferWithExtendableBuffer {
  private:
     DISALLOW_COPY_AND_ASSIGN(BufferWithExtendableBuffer);
 
-    static const size_t INITIAL_ADDITIONAL_BUFFER_SIZE;
     static const size_t MAX_ADDITIONAL_BUFFER_SIZE;
+    static const int NEAR_BUFFER_LIMIT_THRESHOLD_PERCENTILE;
     static const size_t EXTEND_ADDITIONAL_BUFFER_SIZE_STEP;
 
     uint8_t *const mOriginalBuffer;
     const int mOriginalBufferSize;
     std::vector<uint8_t> mAdditionalBuffer;
     int mUsedAdditionalBufferSize;
+    const size_t mMaxAdditionalBufferSize;
 
     // Return if the buffer is successfully extended or not.
-    AK_FORCE_INLINE bool extendBuffer() {
+    bool extendBuffer();
-        if (mAdditionalBuffer.size() + EXTEND_ADDITIONAL_BUFFER_SIZE_STEP
-                > MAX_ADDITIONAL_BUFFER_SIZE) {
-            return false;
-        }
-        mAdditionalBuffer.resize(mAdditionalBuffer.size() + EXTEND_ADDITIONAL_BUFFER_SIZE_STEP);
-        return true;
-    }
 
     // Returns if it is possible to write size-bytes from pos. When pos is at the tail position of
     // the additional buffer, try extending the buffer.
-    AK_FORCE_INLINE bool checkAndPrepareWriting(const int pos, const int size);
+    bool checkAndPrepareWriting(const int pos, const int size);
 };
 }
 #endif /* LATINIME_BUFFER_WITH_EXTENDABLE_BUFFER_H */

tests/src/com/android/inputmethod/latin/BinaryDictionaryTests.java

@@ -604,4 +604,50 @@ public class BinaryDictionaryTests extends AndroidTestCase {
 
         dictFile.delete();
     }
+
+    public void testAddManyUnigramsAndFlushWithGC() {
+        final int flashWithGCIterationCount = 3;
+        final int codePointSetSize = 50;
+        final int seed = 22360679;
+
+        final Random random = new Random(seed);
+
+        File dictFile = null;
+        try {
+            dictFile = createEmptyDictionaryAndGetFile("TestBinaryDictionary");
+        } catch (IOException e) {
+            fail("IOException while writing an initial dictionary : " + e);
+        } catch (UnsupportedFormatException e) {
+            fail("UnsupportedFormatException while writing an initial dictionary : " + e);
+        }
+
+        final ArrayList<String> words = new ArrayList<String>();
+        final HashMap<String, Integer> unigramProbabilities = new HashMap<String, Integer>();
+        final int[] codePointSet = CodePointUtils.generateCodePointSet(codePointSetSize, random);
+
+        BinaryDictionary binaryDictionary;
+        for (int i = 0; i < flashWithGCIterationCount; i++) {
+            binaryDictionary = new BinaryDictionary(dictFile.getAbsolutePath(),
+                    0 /* offset */, dictFile.length(), true /* useFullEditDistance */,
+                    Locale.getDefault(), TEST_LOCALE, true /* isUpdatable */);
+            while(!binaryDictionary.needsToRunGC()) {
+                final String word = CodePointUtils.generateWord(random, codePointSet);
+                words.add(word);
+                final int unigramProbability = random.nextInt(0xFF);
+                unigramProbabilities.put(word, unigramProbability);
+                binaryDictionary.addUnigramWord(word, unigramProbability);
+            }
+
+            for (int j = 0; j < words.size(); j++) {
+                final String word = words.get(j);
+                final int unigramProbability = unigramProbabilities.get(word);
+                assertEquals(word, unigramProbability, binaryDictionary.getFrequency(word));
+            }
+
+            binaryDictionary.flushWithGC();
+            binaryDictionary.close();
+        }
+
+        dictFile.delete();
+    }
 }