Check url path as well as document title to detect formless autofill page

components/autofill/content/renderer/form_autofill_util.cc

 // Copyright 2013 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.
 
 #include "components/autofill/content/renderer/form_autofill_util.h"
 
 #include <map>
 #include <set>
 
 #include "base/command_line.h"
@@ skipping 1386 matching lines @@
 bool UnownedCheckoutFormElementsAndFieldSetsToFormData(
     const std::vector<blink::WebElement>& fieldsets,
     const std::vector<blink::WebFormControlElement>& control_elements,
     const blink::WebFormControlElement* element,
     const blink::WebDocument& document,
     ExtractMask extract_mask,
     FormData* form,
     FormFieldData* field) {
   // Only attempt formless Autofill on checkout flows. This avoids the many
   // false positives found on the non-checkout web. See
-  // http://crbug.com/462375. For now this early abort only applies to
-  // English-language pages, because the regex is not translated. Note that
-  // an empty "lang" attribute counts as English. A potential problem is that
-  // this only checks document.title(), but should actually check the main
-  // frame's title. Thus it may make bad decisions for iframes.
+  // http://crbug.com/462375.
   WebElement html_element = document.documentElement();

[jungshik at Google, 2015/12/16 21:14:04]

Although not very common, there are documents with different sections tagged
with different languages. If we know the form element to deal with, we should
get 'lang' corresponding to that element instead of 'html' element. 

+
+  // For now this restriction only applies to English-language pages, because
+  // the keywords are not translated. Note that an empty "lang" attribute
+  // counts as English.

[jungshik at Google, 2015/12/14 23:53:56]

It'd better to treat an empty 'lang' as in the current UI language. 
In the future,  we may also consider detecting the language (using CLD(2)) .  If
translation feature is on, the language detection must have been run (already) .
If that can be retrieved here, it'd be nice. 

   std::string lang;
   if (!html_element.isNull())
     lang = html_element.getAttribute("lang").utf8();
-  if (lang.empty() ||
-      base::StartsWith(lang, "en", base::CompareCase::INSENSITIVE_ASCII)) {
-    std::string title(base::UTF16ToUTF8(base::string16(document.title())));
-    const char* const kKeywords[] = {
-      "payment",
-      "checkout",
-      "address",
-      "delivery",
-      "shipping",
-    };
+  if (!lang.empty() &&
+      !base::StartsWith(lang, "en", base::CompareCase::INSENSITIVE_ASCII)) {

[jungshik at Google, 2015/12/14 23:53:57]

This assumes that there is no language code (3-letter) that starts wtih 'en'
(other than English). 
To be more robust, either use ui/base/l10n_util.h (GetLanguage) or use ICU's
locale API. 

+    return UnownedFormElementsAndFieldSetsToFormData(
+        fieldsets, control_elements, element, document, extract_mask, form,
+        field);
+  }
 
-    bool found = false;
-    for (const auto& keyword : kKeywords) {
-      if (title.find(keyword) != base::string16::npos) {
-        found = true;
-        break;
-      }
+  // A potential problem is that this only checks document.title(), but should
+  // actually check the main frame's title. Thus it may make bad decisions for
+  // iframes.
+  base::string16 title(base::ToLowerASCII(base::string16(document.title())));
+
+  // Don't check the path for url's without a standard format path component,
+  // such as data:.
+  std::string path;
+  GURL url(document.url());
+  if (url.IsStandard())
+    path = base::ToLowerASCII(url.path());
+
+  const char* const kKeywords[] = {
+    "payment",
+    "checkout",
+    "address",
+    "delivery",
+    "shipping",
+  };
+
+  bool found = false;
+  for (const auto& keyword : kKeywords) {
+    // Compare char16 elements of |title| with char elements of |keyword| using
+    // operator==.
+    auto title_pos = std::search(title.begin(), title.end(),
+                                 keyword, keyword + strlen(keyword));
+    if (title_pos != title.end() ||
+        path.find(keyword) != std::string::npos) {
+      found = true;
+      break;
     }
-    if (!found)
-      return false;
   }
+  if (!found)
+    return false;
 
   return UnownedFormElementsAndFieldSetsToFormData(

[Evan Stade, 2016/01/08 21:53:16]

seems like you can just stick this inside the loop (where you have found = true)
and get rid of |found|

       fieldsets, control_elements, element, document, extract_mask, form,
       field);
 }
 
 bool UnownedPasswordFormElementsAndFieldSetsToFormData(
     const std::vector<blink::WebElement>& fieldsets,
     const std::vector<blink::WebFormControlElement>& control_elements,
     const blink::WebFormControlElement* element,
     const blink::WebDocument& document,
     ExtractMask extract_mask,
@@ skipping 196 matching lines @@
     // Zero selection start is for password manager, which can show usernames
     // that do not begin with the user input value.
     selection_start = (offset == base::string16::npos) ? 0 : offset;
   }
 
   input_element->setSelectionRange(selection_start, suggestion.length());
 }
 
 }  // namespace form_util
 }  // namespace autofill