Language attribute should only contain first value from meta tag.

AXNodeObject::Language was returning the full meta tag value "en, fr"
instead of only the first element "en" when computing the language value
for kRootWebArea.

A value containing comma is not a valid BCP47 language tag.

[1] specifies the DOM lang attribute should be a valid BCP47 language
tag, whereas a comma separated list is instead a language priority list.
I'm not sure if this necessarily applies to our accessible interface,
but it seems like the closest spec and this behaviour is consistent
with our other return values.

[1] https://html.spec.whatwg.org/multipage/dom.html#language

AX-Relnotes: n/a.

Change-Id: I0e7b2826e3cca0b9392a4add39c99244018e9dfb
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2505291
Commit-Queue: Chris Hall <chrishall@chromium.org>
Reviewed-by: Alice Boxhall <aboxhall@chromium.org>
Reviewed-by: Nektarios Paisios <nektar@chromium.org>
Cr-Commit-Position: refs/heads/master@{#825263}

third_party/blink/renderer/modules/accessibility/ax_node_object.cc

@@ -3639,6 +3639,8 @@ AtomicString AXNodeObject::Language() const {
   if (!GetNode())
     return AXObject::Language();
 
+  Vector<String> languages;
+
   // If it's the root, get the computed language for the document element,
   // because the root LayoutObject doesn't have the right value.
   if (RoleValue() == ax::mojom::blink::Role::kRootWebArea) {
@@ -3646,9 +3648,16 @@ AtomicString AXNodeObject::Language() const {
     if (!document_element)
       return g_empty_atom;
 
+    // Ensure we return only the first language tag. ComputeInheritedLanguage
+    // consults ContentLanguage which can be set from 2 different sources.
+    // DocumentLoader::DidInstallNewDocument from HTTP headers which truncates
+    // until the first comma.
+    // HttpEquiv::Process from <meta> tag which does not truncate.
+    // TODO(chrishall): Consider moving this comma handling to setter side.
     AtomicString lang = document_element->ComputeInheritedLanguage();
-    if (!lang.IsEmpty())
-      return lang;
+    String(lang).Split(',', languages);
+    if (!languages.IsEmpty())
+      return AtomicString(languages[0].StripWhiteSpace());
   }
 
   // Uses the style engine to figure out the object's language.
@@ -3660,7 +3669,6 @@ AtomicString AXNodeObject::Language() const {
   if (!style || !style->Locale())
     return AXObject::Language();
 
-  Vector<String> languages;
   String(style->Locale()).Split(',', languages);
   if (languages.IsEmpty())
     return AXObject::Language();

third_party/blink/web_tests/accessibility/language-in-canvas.html

@@ -15,6 +15,7 @@
 
   <script>
     setup(() => {
+      window.axRoot = accessibilityController.rootElement;
       window.axCanvas = accessibilityController.accessibleElementById('canvas');
       window.axParagraph = accessibilityController.accessibleElementById('paragraph');
       window.axSpan = accessibilityController.accessibleElementById('span');
@@ -23,6 +24,8 @@
     async_test((t) => {
       internals.setUserPreferredLanguages([ 'de' ]);
 
+      assert_equals(window.axRoot.language, 'AXLanguage: en',
+          'The first language in the meta tag should apply.');
       assert_equals(window.axCanvas.language, 'AXLanguage: en',
           'The first language in the meta tag should apply.');
       assert_equals(window.axParagraph.language, 'AXLanguage: en',
@@ -35,6 +38,8 @@
       meta.content = '';
 
       t.step_timeout(() => {
+        assert_equals(window.axRoot.language, 'AXLanguage: de',
+            'The browser language should apply.');
         assert_equals(window.axCanvas.language, 'AXLanguage: de',
             'The browser language should apply.');
         assert_equals(window.axParagraph.language, 'AXLanguage: de',