[Password Manager] Polishing HTML based Username Detector

Bug: 699530
Change-Id: I6325ed2be5868d519c5b0d0cb720d09e7b80da4d
Reviewed-on: https://chromium-review.googlesource.com/759792
Commit-Queue: Maxim Kolosovskiy <kolos@chromium.org>
Reviewed-by: Vaclav Brozek <vabr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#521018}

components/autofill/content/renderer/BUILD.gn

@@ -14,6 +14,8 @@ static_library("renderer") {
     "form_classifier.h",
     "html_based_username_detector.cc",
     "html_based_username_detector.h",
+    "html_based_username_detector_vocabulary.cc",
+    "html_based_username_detector_vocabulary.h",
     "page_form_analyser_logger.cc",
     "page_form_analyser_logger.h",
     "page_passwords_analyser.cc",

components/autofill/content/renderer/html_based_username_detector.h

@@ -2,6 +2,8 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#include <map>
+
 #include "components/autofill/core/common/password_form.h"
 #include "third_party/WebKit/public/web/WebFormControlElement.h"
 #include "third_party/WebKit/public/web/WebInputElement.h"

components/autofill/content/renderer/html_based_username_detector_vocabulary.h

+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+namespace autofill {
+
+// Words that certainly point to a non-username field.
+// If field values contain at least one negative word, then the field is
+// excluded from the list of possible usernames.
+extern const char* const kNegativeLatin[];
+extern const int kNegativeLatinSize;
+extern const char* const kNegativeNonLatin[];
+extern const int kNegativeNonLatinSize;
+
+// Translations of "username".
+extern const char* const kUsernameLatin[];
+extern const int kUsernameLatinSize;
+extern const char* const kUsernameNonLatin[];
+extern const int kUsernameNonLatinSize;
+
+// Translations of "user".
+extern const char* const kUserLatin[];
+extern const int kUserLatinSize;
+extern const char* const kUserNonLatin[];
+extern const int kUserNonLatinSize;
+
+// Words that certainly point to a username field, if they appear in developer
+// value. They are technical words, because they can only be used as variable
+// names, and not as stand-alone words.
+extern const char* const kTechnicalWords[];
+extern const int kTechnicalWordsSize;
+
+// Words that might point to a username field.They have the smallest priority
+// in the heuristic, because there are also field attribute values that
+// contain them, but are not username fields.
+extern const char* const kWeakWords[];
+extern const int kWeakWordsSize;
+
+}  // namespace autofill

components/autofill/content/renderer/password_form_conversion_utils_browsertest.cc

@@ -221,7 +221,8 @@ class MAYBE_PasswordFormConversionUtilsTest : public content::RenderViewTest {
     }
 
     return CreatePasswordFormFromWebForm(
-        form, with_user_input ? &user_input : nullptr, predictions, nullptr);
+        form, with_user_input ? &user_input : nullptr, predictions,
+        &username_detector_cache_);
   }
 
   // Iterates on the form generated by the |html| and adds the fields and type
@@ -270,6 +271,8 @@ class MAYBE_PasswordFormConversionUtilsTest : public content::RenderViewTest {
     *form = forms[0];
   }
 
+  UsernameDetectorCache username_detector_cache_;
+
  private:
   DISALLOW_COPY_AND_ASSIGN(MAYBE_PasswordFormConversionUtilsTest);
 };
@@ -334,7 +337,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest, OnlyDisabledFields) {
 }
 
 TEST_F(MAYBE_PasswordFormConversionUtilsTest,
-       IdentifyingUsernameFieldsFromDeveloperGroupWithHTMLDetector) {
+       HTMLDetector_DeveloperGroupAttributes) {
   base::test::ScopedFeatureList feature_list;
   feature_list.InitAndEnableFeature(
       password_manager::features::kEnableHtmlBasedUsernameDetector);
@@ -408,17 +411,18 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
        {"email", "", "js@google.com"},
        "email",
        "js@google.com"},
-      // If word matches in maximum 2 fields, it is accepted.
+      // If a word matches in maximum 2 fields, it is accepted.
       // First encounter is selected as username.
-      {{"loginusername", "", "johnsmith"},
-       {"loginemail", "", "js@google.com"},
-       "loginusername",
+      {{"username", "", "johnsmith"},
+       {"repeat_username", "", "johnsmith"},
+       "username",
        "johnsmith"},
-      // Check treatment for short dictionary words.
-      {{"identity_name", "", "johnsmith"},
-       {"email", "", "js@google.com"},
-       "email",
-       "js@google.com"}};
+      // A short word should be enclosed between delimiters. Otherwise, an
+      // Occurrence doesn't count.
+      {{"identity_name", "idn", "johnsmith"},
+       {"id", "id", "123"},
+       "id",
+       "123"}};
 
   for (size_t i = 0; i < arraysize(cases); ++i) {
     SCOPED_TRACE(testing::Message() << "Iteration " << i);
@@ -436,6 +440,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
     builder.AddSubmitButton("submit");
     std::string html = builder.ProduceHTML();
 
+    username_detector_cache_.clear();
     std::unique_ptr<PasswordForm> password_form =
         LoadHTMLAndConvertForm(html, nullptr, false);
 
@@ -445,7 +450,19 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
               password_form->username_element);
     EXPECT_EQ(base::UTF8ToUTF16(cases[i].expected_username_value),
               password_form->username_value);
+    // Check that the username field was found by HTML detector.
+    ASSERT_EQ(1u, username_detector_cache_.size());
+    ASSERT_FALSE(username_detector_cache_.begin()->second.IsNull());
+    EXPECT_EQ(
+        cases[i].expected_username_element,
+        username_detector_cache_.begin()->second.NameForAutofill().Utf8());
   }
+}
+
+TEST_F(MAYBE_PasswordFormConversionUtilsTest, HTMLDetector_SeveralDetections) {
+  base::test::ScopedFeatureList feature_list;
+  feature_list.InitAndEnableFeature(
+      password_manager::features::kEnableHtmlBasedUsernameDetector);
 
   // If word matches in more than 2 fields, we don't match on it.
   // We search for match with another word.
@@ -459,6 +476,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
   builder.AddSubmitButton("submit");
   std::string html = builder.ProduceHTML();
 
+  DCHECK(username_detector_cache_.empty());
   std::unique_ptr<PasswordForm> password_form =
       LoadHTMLAndConvertForm(html, nullptr, false);
 
@@ -466,10 +484,15 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
 
   EXPECT_EQ(base::UTF8ToUTF16("loginid"), password_form->username_element);
   EXPECT_EQ(base::UTF8ToUTF16("johnsmith"), password_form->username_value);
+  // Check that the username field was found by HTML detector.
+  ASSERT_EQ(1u, username_detector_cache_.size());
+  ASSERT_FALSE(username_detector_cache_.begin()->second.IsNull());
+  EXPECT_EQ("loginid",
+            username_detector_cache_.begin()->second.NameForAutofill().Utf8());
 }
 
 TEST_F(MAYBE_PasswordFormConversionUtilsTest,
-       IdentifyingUsernameFieldsFromUserGroupWithHTMLDetector) {
+       HTMLDetector_UserGroupAttributes) {
   base::test::ScopedFeatureList feature_list;
   feature_list.InitAndEnableFeature(
       password_manager::features::kEnableHtmlBasedUsernameDetector);
@@ -483,12 +506,12 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
   struct TestCase {
     // Field parameters represent, in order of appearance, field name, field
     // id, field value and field label or placeholder.
+    // Field name and field id don't contain any significant information.
     const char* first_text_field_parameters[4];
     const char* second_text_field_parameters[4];
     const char* expected_username_element;
     const char* expected_username_value;
   } cases[] = {
-      // Developer group does not contain any significant information.
       // Label information will decide username.
       {{"name1", "id1", "johnsmith", "Username:"},
        {"name2", "id2", "js@google.com", "Email:"},
@@ -545,9 +568,10 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
        {"username", "", "johnsmith", "Email:"},
        "email",
        "js@google.com"},
-      // Check treatment for short dictionary words.
+      // Check treatment for short dictionary words. "uid" has higher priority,
+      // but its occurrence is ignored because it is a part of another word.
       {{"name1", "", "johnsmith", "Insert your id:"},
-       {"name2", "", "js@google.com", "Insert something:"},
+       {"name2", "uidentical", "js@google.com", "Insert something:"},
        "name1",
        "johnsmith"}};
 
@@ -569,6 +593,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
     builder.AddSubmitButton("submit");
     std::string html = builder.ProduceHTML();
 
+    username_detector_cache_.clear();
     std::unique_ptr<PasswordForm> password_form =
         LoadHTMLAndConvertForm(html, nullptr, false);
 
@@ -578,6 +603,12 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest,
               password_form->username_element);
     EXPECT_EQ(base::UTF8ToUTF16(cases[i].expected_username_value),
               password_form->username_value);
+    // Check that the username field was found by HTML detector.
+    ASSERT_EQ(1u, username_detector_cache_.size());
+    ASSERT_FALSE(username_detector_cache_.begin()->second.IsNull());
+    EXPECT_EQ(
+        cases[i].expected_username_element,
+        username_detector_cache_.begin()->second.NameForAutofill().Utf8());
   }
 }
 
@@ -613,7 +644,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest, HTMLDetectorCache) {
   // will be the same because it was cached in |username_detector_cache|.
   WebVector<WebFormControlElement> control_elements;
   form.GetFormControlElements(control_elements);
-  control_elements[0].SetAttribute("name", "login");
+  control_elements[0].SetAttribute("name", "id");
   password_form = CreatePasswordFormFromWebForm(form, nullptr, nullptr,
                                                 &username_detector_cache);
   EXPECT_TRUE(password_form);
@@ -633,7 +664,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest, HTMLDetectorCache) {
   ASSERT_EQ(1u, username_detector_cache.size());
   EXPECT_EQ(form, username_detector_cache.begin()->first);
   ASSERT_FALSE(username_detector_cache.begin()->second.IsNull());
-  EXPECT_EQ("login",
+  EXPECT_EQ("id",
             username_detector_cache.begin()->second.NameForAutofill().Utf8());
   EXPECT_THAT(
       histogram_tester.GetAllSamples("PasswordManager.UsernameDetectionMethod"),
@@ -650,7 +681,7 @@ TEST_F(MAYBE_PasswordFormConversionUtilsTest, HTMLDetectorCache) {
   ASSERT_EQ(1u, username_detector_cache.size());
   EXPECT_EQ(form, username_detector_cache.begin()->first);
   ASSERT_FALSE(username_detector_cache.begin()->second.IsNull());
-  EXPECT_EQ("login",
+  EXPECT_EQ("id",
             username_detector_cache.begin()->second.NameForAutofill().Utf8());
   EXPECT_THAT(
       histogram_tester.GetAllSamples("PasswordManager.UsernameDetectionMethod"),