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

components/autofill/content/renderer/form_autofill_util.cc

Index: components/autofill/content/renderer/form_autofill_util.cc
diff --git a/components/autofill/content/renderer/form_autofill_util.cc b/components/autofill/content/renderer/form_autofill_util.cc
index 5c7f0d6102e281c6fa8f457f2e2e7e5acb38a5be..57d989d75773492d9aced89c3610db9dc2ef0a02 100644
--- a/components/autofill/content/renderer/form_autofill_util.cc
+++ b/components/autofill/content/renderer/form_autofill_util.cc
@@ -1404,36 +1404,53 @@ bool UnownedCheckoutFormElementsAndFieldSetsToFormData(
     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();
+
+  // 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.
   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",
-    };
-
-    bool found = false;
-    for (const auto& keyword : kKeywords) {
-      if (title.find(keyword) != base::string16::npos) {
-        found = true;
-        break;
-      }
+  if (!lang.empty() &&
+      !base::StartsWith(lang, "en", base::CompareCase::INSENSITIVE_ASCII)) {
+    return UnownedFormElementsAndFieldSetsToFormData(
+        fieldsets, control_elements, element, document, extract_mask, form,
+        field);
+  }
+
+  // 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.
+  std::string title = base::ToLowerASCII(base::UTF16ToUTF8(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) {
+    if (title.find(keyword) != base::string16::npos ||

[Evan Stade, 2015/12/10 19:34:00]

I think you should use std::find_if with base::LowerCaseEqualsASCII

[Joe Mason, 2015/12/10 20:06:58]

That would be less efficient since we need to convert title each time through
the loop. It probably doesn't make much difference, but what's the benefit?

[Evan Stade, 2015/12/10 21:21:11]

You're right, probably not a huge difference, but fewer lines of code and no
utf16->utf8 conversion (which is much more costly than Z->z).

[Joe Mason, 2015/12/11 20:41:43]

Done.

+        path.find(keyword) != base::string16::npos) {
+      found = true;
+      break;
     }
-    if (!found)
-      return false;
   }
+  if (!found)
+    return false;
 
   return UnownedFormElementsAndFieldSetsToFormData(
       fieldsets, control_elements, element, document, extract_mask, form,