Support accessible inline text boxes.

content/browser/accessibility/browser_accessibility_win.cc

Index: content/browser/accessibility/browser_accessibility_win.cc
diff --git a/content/browser/accessibility/browser_accessibility_win.cc b/content/browser/accessibility/browser_accessibility_win.cc
index df3263e845ddfb2cc82fe58cb5ce1b117d08bcda..9dc644c9423bd8d2eab80b8a9654343859b7af6a 100644
--- a/content/browser/accessibility/browser_accessibility_win.cc
+++ b/content/browser/accessibility/browser_accessibility_win.cc
@@ -358,7 +358,11 @@ STDMETHODIMP BrowserAccessibilityWin::get_accChildCount(LONG* child_count) {
   if (!child_count)
     return E_INVALIDARG;
 
-  *child_count = children_.size();
+  if (IsLeaf())
+    *child_count = 0;
+  else
+    *child_count = children_.size();
+
   return S_OK;
 }
 
@@ -2618,7 +2622,7 @@ STDMETHODIMP BrowserAccessibilityWin::get_childAt(
   if (!node)
     return E_INVALIDARG;
 
-  if (child_index < children_.size()) {
+  if (IsLeaf() || child_index < children_.size()) {

[David Tseng, 2013/10/23 16:42:46]

I'm confused by this indexing logic;  child_index < child_.size()? So, to index
into children_, child_index needs to be >= children_.size()?

[dmazzoni, 2013/10/30 16:45:26]

Ack! Yes, you found a longstanding bug. We never caught it because it only
affected ISimpleDOMNode. I think ISimpleDOMNode is used occasionally, but no
screen reader appears to ever use it to walk the DOM tree.

I replaced this with PlatformGetChild followed by a NULL check.

     *node = NULL;
     return S_FALSE;
   }