As per discussion, patch attached.

Comment on attachment 486414 [details] [diff] [review] Patch, v1 Make the event go to the transaction after the request. Remove the new and existing event stuff from cursor/index/objectStore. Remove onsuccess (function and member) from transaction/database. Remove onblocked (function and member) from database. Blocked is fired at the request, no? In MaybeDuplicate, don't copy the 'cancelable' flag. It should always be set to false. Cancelling the error event prevents rolling back when done on the request/transaction/database. However the error on the window is purely a error-reporting mechanism. >diff --git a/dom/indexedDB/IDBEvents.cpp b/dom/indexedDB/IDBEvents.cpp > IDBEvent::CreateGenericEvent(const nsAString& aType) > { > nsRefPtr<nsDOMEvent> event(new nsDOMEvent(nsnull, nsnull)); >- nsresult rv = event->InitEvent(aType, PR_FALSE, PR_FALSE); >+ nsresult rv = event->InitEvent(aType, PR_TRUE, PR_TRUE); Why TRUE/TRUE? I suspect this should remain FALSE/FALSE. >@@ -213,18 +273,18 @@ IDBSuccessEvent::Create(IDBRequest* aReq >- nsresult rv = event->InitEvent(NS_LITERAL_STRING(SUCCESS_EVT_STR), PR_FALSE, >- PR_FALSE); >+ nsresult rv = event->InitEvent(NS_LITERAL_STRING(SUCCESS_EVT_STR), PR_TRUE, >+ PR_TRUE); FALSE/FALSE here too. >@@ -306,18 +366,18 @@ GetSuccessEvent::~GetSuccessEvent() >- nsresult rv = InitEvent(NS_LITERAL_STRING(SUCCESS_EVT_STR), PR_FALSE, >- PR_FALSE); >+ nsresult rv = InitEvent(NS_LITERAL_STRING(SUCCESS_EVT_STR), PR_TRUE, >+ PR_TRUE); FALSE/FALSE >@@ -520,17 +580,17 @@ IDBVersionChangeEvent::CreateInternal(ns >- nsresult rv = event->InitEvent(aType, PR_FALSE, PR_FALSE); >+ nsresult rv = event->InitEvent(aType, PR_TRUE, PR_TRUE); FALSE/FALSE r=me with that fixed.

http://hg.mozilla.org/mozilla-central/rev/4211f1d517ed The test was being a little flaky, I also landed these, hopefully they fix it: http://hg.mozilla.org/mozilla-central/rev/bc78b38106e8 http://hg.mozilla.org/mozilla-central/rev/d7f4a4464ea1 http://hg.mozilla.org/mozilla-central/rev/9d15bba46bee