I was reported about http://www.freedomscientific.com. Hitting reload doesn't make a focus. The problem here we receive DOM focus event for document a little bit early than anything is loaded into the document. Obviously this is timing issue but due to some reason it happens on fs site permanently (I'm not aware of other examples). We don't create accessible document (because of bug 612830) and therefore we miss the focus event. Fixing bug 612830 should fix this one.

Comment on attachment 669171 [details] [diff] [review] add logging reveling the problem > LogPresShell(nsIDocument* aDocumentNode) > { > nsIPresShell* ps = aDocumentNode->GetShell(); >- printf("presshell: %p", static_cast<void*>(ps)); >+ printf("presshell: %p, %s destroying", static_cast<void*>(ps), >+ (ps->IsDestroying() ? "" : "not")); maybe "is" instead of "" to be more clear >+logging::Document(DocAccessible* aDocument) >+{ >+ printf(" Document: %p, document node: %p\n", >+ static_cast<void*>(aDocument), >+ static_cast<void*>(aDocument->GetDocumentNode())); >+ >+ printf(" Document "); >+ LogDocURI(static_cast<nsIDocument*>(aDocument->GetDocumentNode())); why is cast needed? > logging::AccessibleNNode(const char* aDescr, nsINode* aNode) > { > DocAccessible* document = > GetAccService()->GetDocAccessible(aNode->OwnerDoc()); > > if (document) { > Accessible* accessible = document->GetAccessible(aNode); > if (accessible) { > AccessibleNNode(aDescr, accessible); > return; > } > } > >- nsAutoCString nodeDescr("Not accessible "); >+ nsAutoCString nodeDescr("[not accessible] "); > nodeDescr.Append(aDescr); > Node(nodeDescr.get(), aNode); >+ >+ if (document) { >+ Document(document); >+ return; >+ } >+ >+ printf(" [contained by document]:\n"); non accessible document?

landed with Trevor's comments fixed http://hg.mozilla.org/integration/mozilla-inbound/rev/b7d112dee89c

it's no longer an issue after bug 612830 was fixed.