Workaround a crash occuring only in v20 and v21 without a repro case (need more data for good fix)

content/browser/renderer_host/render_widget_host_impl.cc

Index: content/browser/renderer_host/render_widget_host_impl.cc
diff --git a/content/browser/renderer_host/render_widget_host_impl.cc b/content/browser/renderer_host/render_widget_host_impl.cc
index e8b5dcbc5b6e19f672fe80df2a1df6617068b40d..cf38f1dfe4e1efdca72eda132e41f6adb5626f30 100644
--- a/content/browser/renderer_host/render_widget_host_impl.cc
+++ b/content/browser/renderer_host/render_widget_host_impl.cc
@@ -880,7 +880,11 @@ void RenderWidgetHostImpl::ForwardKeyboardEvent(
 
     bool is_keyboard_shortcut = false;
     // Only pre-handle the key event if it's not handled by the input method.
-    if (!key_event.skip_in_browser) {
+    // Looking extensively for over a day, a delegate_ of NULL seems

[brettw, 2012/07/23 17:44:45]

This comment isn't clear when not in the context of the diff. It's not clear
what the "fix" is that it's referring to. Also, I'm not sure if "looking
extensively for over a day" is necessary.

I'd write something like "A delegate_ of NULL seems impossible but crash reports
show that it can happen (see http://crbug.com/134465). This doesn't seem to
happen in Chrome 22 and later, so checking for the delegate_ here can be removed
once Chrome 22 goes to stable."

[Mr4D (OOO till 08-26), 2012/07/23 21:46:17]

Done.

+    // impossible. But (crbug/134465) has shown that it can happen in some
+    // non reproducible cases. With v22 this problem did not show anymore so
+    // that we can get rid of this 'fix' later.
+    if (delegate_ && !key_event.skip_in_browser) {
       // We need to set |suppress_next_char_events_| to true if
       // PreHandleKeyboardEvent() returns true, but |this| may already be
       // destroyed at that time. So set |suppress_next_char_events_| true here,