Optimize phishing page term feature extraction.

chrome/renderer/safe_browsing/phishing_term_feature_extractor.h

Index: chrome/renderer/safe_browsing/phishing_term_feature_extractor.h
diff --git a/chrome/renderer/safe_browsing/phishing_term_feature_extractor.h b/chrome/renderer/safe_browsing/phishing_term_feature_extractor.h
index 74c9b0b9acec0dbc939a69370caca08331c37bf6..384e093cdb9de1ab48725697c66e85f2ddef7416 100644
--- a/chrome/renderer/safe_browsing/phishing_term_feature_extractor.h
+++ b/chrome/renderer/safe_browsing/phishing_term_feature_extractor.h
@@ -21,9 +21,11 @@
 #include "base/basictypes.h"
 #include "base/callback_old.h"
 #include "base/hash_tables.h"
+#include "base/memory/mru_cache.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/string16.h"
 #include "base/task.h"
+#include "base/wide_string_piece.h"
 
 namespace safe_browsing {
 class FeatureExtractorClock;
@@ -99,7 +101,7 @@ class PhishingTermFeatureExtractor {
   void ExtractFeaturesWithTimeout();
 
   // Handles a single word in the page text.
-  void HandleWord(const string16& word);
+  void HandleWord(const base::WideStringPiece& word);
 
   // Helper to verify that there is no pending feature extraction.  Dies in
   // debug builds if the state is not as expected.  This is a no-op in release
@@ -125,6 +127,13 @@ class PhishingTermFeatureExtractor {
   // The maximum number of words in an n-gram.
   size_t max_words_per_term_;
 
+  // This cache is used to see if we need to check the word at all, as
+  // converting to UTF8, lowercasing, and hashing are all relatively expensive
+  // operations. Though this is called an MRU cache, it seems to behave like
+  // an LRU cache (i.e. it evicts the oldest accesses first).
+  typedef base::HashingMRUCache<base::WideStringPiece, bool> WordCache;
+  WordCache negative_word_cache_;
+
   // Non-owned pointer to our clock.
   FeatureExtractorClock* clock_;