Followup to transfer serviceworker state during cross site navigations too.

content/browser/service_worker/service_worker_provider_host.cc

Index: content/browser/service_worker/service_worker_provider_host.cc
diff --git a/content/browser/service_worker/service_worker_provider_host.cc b/content/browser/service_worker/service_worker_provider_host.cc
index 2bb6499487916db409c6eecbd9c10cae0863b7e2..0de2aa243f73ecb3b4e79533c359ecf38e50f239 100644
--- a/content/browser/service_worker/service_worker_provider_host.cc
+++ b/content/browser/service_worker/service_worker_provider_host.cc
@@ -258,8 +258,13 @@ void ServiceWorkerProviderHost::PrepareForCrossSiteTransfer() {
   for (const GURL& pattern : associated_patterns_)
     DecreaseProcessReference(pattern);
 
-  if (associated_registration_.get())
+  if (associated_registration_.get()) {
     DecreaseProcessReference(associated_registration_->pattern());
+    if (dispatcher_host_) {
+      dispatcher_host_->Send(new ServiceWorkerMsg_DisassociateRegistration(

[michaeln, 2014/12/12 01:22:17]

This is what i had messed up, not disassociating things in the renderer side.

In the context of the browser test, a new processes is not actually started, so
the 'host' gets transferred back into the same process it originally came out
of. But the renderer side asserts that there is no registration when its
handling a ServiceWorkerMsg_AssociateRegistration. That message gets sent inside
of CompleteCrossSiteTransfer and the DCHECK in the renderer was causing the test
to fail.

So we disassociate here first so that we can safely CompleteCrossSiteTransfer
even to the same process.

+          kDocumentMainThreadId, provider_id()));
+    }
+  }
 
   render_process_id_ = ChildProcessHost::kInvalidUniqueID;
   render_frame_id_ = MSG_ROUTING_NONE;