Reset accessibility if it gets out of sync.

content/common/accessibility_messages.h

Index: content/common/accessibility_messages.h
diff --git a/content/common/accessibility_messages.h b/content/common/accessibility_messages.h
index 02af80708cce2f135c17b19427b939521a71fab6..f5ef3c50f8d3c822bacb7bdc57485e600e67087e 100644
--- a/content/common/accessibility_messages.h
+++ b/content/common/accessibility_messages.h
@@ -131,16 +131,30 @@ IPC_MESSAGE_ROUTED1(AccessibilityMsg_HitTest,
 // message was processed and it can send addition events.
 IPC_MESSAGE_ROUTED0(AccessibilityMsg_Events_ACK)
 
-// Kill the renderer because we got a fatal error in the accessibility tree.
+// Tell the renderer to reset and send a new accessibility tree from
+// scratch because the browser is out of sync. This should be rare, and
+// if we need reset the same renderer too many times we just kill it.
+// After sending a reset, the browser ignores incoming accessibility IPCs
+// until it receives one with is_reset=true. Conversely, it ignores IPCs
+// with is_reset=true if it was not expecting a reset.
+IPC_MESSAGE_ROUTED0(AccessibilityMsg_Reset)
+
+// Kill the renderer because we got a fatal error in the accessibility tree
+// and we've already reset too many times.
 IPC_MESSAGE_ROUTED0(AccessibilityMsg_FatalError)
 
 // Messages sent from the renderer to the browser.
 
 // Sent to notify the browser about renderer accessibility events.
 // The browser responds with a AccessibilityMsg_Events_ACK.
-IPC_MESSAGE_ROUTED1(
+// The second parameter, is_reset, is set if this IPC was sent in response
+// to a reset request from the browser. When the browser requests a reset,
+// it ignores incoming IPCs until it sees one with is_reset=true. Any other
+// time, it ignores IPCs with is_reset=true.
+IPC_MESSAGE_ROUTED2(
     AccessibilityHostMsg_Events,
-    std::vector<AccessibilityHostMsg_EventParams>)
+    std::vector<AccessibilityHostMsg_EventParams> /* events */,
+    bool /* is_reset */)

[Tom Sepez, 2014/10/02 16:50:49]

Are you sure you want a bool instead of a reset sequence number?  I could
imagine two resets overlapping somehow -- in other words, you've still got
implied shared state between the renderer and browser that you could make
explicit.

[dmazzoni, 2014/10/02 21:51:13]

I was debating this. I went ahead and gave it a random reset token rather than a
sequence that could easily be guessed, just to be safe.

 
 // Sent to update the browser of the location of accessibility objects.
 IPC_MESSAGE_ROUTED1(