Import libtextclassifier to fix thread safety issue in LangID

BUG: 130153160

Test: atest frameworks/base/core/tests/coretests/src/android/view/textclassifier/

Change-Id: I76c125a026e572cfde941372620450d6c5c3e7f3

3 files changed, 12 insertions, 1 deletions

diff --git a/lang_id/features/char-ngram-feature.cc b/lang_id/features/char-ngram-feature.cc
index e52b2f2..83d7588 100644
--- a/lang_id/features/char-ngram-feature.cc
+++ b/lang_id/features/char-ngram-feature.cc
@@ -123,6 +123,11 @@ int ContinuousBagOfNgramsFunction::ComputeNgramCounts(
 void ContinuousBagOfNgramsFunction::Evaluate(const WorkspaceSet &workspaces,
                                              const LightSentence &sentence,
                                              FeatureVector *result) const {
+  // NOTE: we use std::* constructs (instead of absl::Mutex & co) to simplify
+  // porting to Android and to avoid pulling in absl (which increases our code
+  // size).
+  std::lock_guard<std::mutex> mlock(state_mutex_);
+
   // Find the char ngram counts.
   int total_count = ComputeNgramCounts(sentence);
 
diff --git a/lang_id/features/char-ngram-feature.h b/lang_id/features/char-ngram-feature.h
index 8280bca..db0f83e 100644
--- a/lang_id/features/char-ngram-feature.h
+++ b/lang_id/features/char-ngram-feature.h
@@ -17,6 +17,7 @@
 #ifndef NLP_SAFT_COMPONENTS_LANG_ID_MOBILE_FEATURES_CHAR_NGRAM_FEATURE_H_
 #define NLP_SAFT_COMPONENTS_LANG_ID_MOBILE_FEATURES_CHAR_NGRAM_FEATURE_H_
 
+#include <mutex>  // NOLINT: see comments for state_mutex_
 #include <string>
 
 #include "lang_id/common/fel/feature-extractor.h"
@@ -69,6 +70,11 @@ class ContinuousBagOfNgramsFunction : public LightSentenceFeature {
   // below), and returns the total ngram count.
   int ComputeNgramCounts(const LightSentence &sentence) const;
 
+  // Guards counts_ and non_zero_count_indices_.  NOTE: we use std::* constructs
+  // (instead of absl::Mutex & co) to simplify porting to Android and to avoid
+  // pulling in absl (which increases our code size).
+  mutable std::mutex state_mutex_;
+
   // counts_[i] is the count of all ngrams with id i.  Work data for Evaluate().
   // NOTE: we declare this vector as a field, such that its underlying capacity
   // stays allocated in between calls to Evaluate().
diff --git a/lang_id/lang-id.h b/lang_id/lang-id.h
index 848ac6e..94af0c3 100644
--- a/lang_id/lang-id.h
+++ b/lang_id/lang-id.h
@@ -53,7 +53,7 @@ struct LangIdResult {
 // Note: this class does not handle the details of loading the actual model.
 // Those details have been "outsourced" to the ModelProvider class.
 //
-// Note: this class is thread-unsafe.
+// This class is thread safe.
 class LangId {
  public:
   // Standard BCP-47 language code for Unknown/Undetermined language.