Here's my current example (simplified for brevity): interface nsIImapMessageSink : nsISupports { void SetupMsgWriteStream(in nsIFileSpec aFileSpec, in boolean appendDummyEnvelope); } Now from thread A which happens to be an imap connection thread, I call setupMsgWriteStream nsCOMPtr<nsIFileSpec> fileSpec; PRBool addDummyEnvelope = PR_TRUE; rv = msgurl->GetMessageFile(getter_AddRefs(fileSpec)); if (NS_FAILED(rv)) return rv; m_imapMessageSink->SetupMsgWriteStream(fileSpec, addDummyEnvelope); Now, thread A is blocked until the message sink call to setupMsgWriteStream is proxied and completed on thread B (which is the UI thread). Thread B makes a call like the following: SetupMsgWriteStream(nsIFileSpec * aFileSpec, PRBool addDummyEnvelope) { *blah* aFileSpec->GetFileSpec(&fileSpec); *blah* } The call to GetFileSpec turns into a auto-proxy call back to thread A. But events aren't getting pumped on thread A because it is blocked waiting for B to return. Hence the dead lock.

I have disabled this. Re-address after beta

I have a nested event loop know which prevents deadlock. All I have to do is remove the #ifdef's which I previously added. I am going to mark this as fixed, but leave the #ifdefs in until I fix 16773 and coclasses are invented.