The tests enabled in bug 1170937 fatally assert under e10s because of this.

This doesn't look very simple. In particular, I'm not sure how to serialize and deserialize the pipe and callback members. Nathan, do you have any ideas?

This is non-trivial. Kyle spent a lot of time trying to make it work and we eventually did something non-generic for Cache instead. See bug 1093357. Still its something we will need eventually. It may have to be something outside SerializeInputStream(), though. It will likely have to return an actor and the user will have to manage that actor appropriately.

In particular see bug 1093357 comment 35.

(In reply to Ben Kelly [:bkelly] from comment #3) > In particular see bug 1093357 comment 35. I have to admit that I don't understand everything in that comment... But what are we going to do here? Wait for bug 1093357? Or are you suggesting that we should work around the issue somehow?

Well, that comment offered a possible path forward. I can possibly punt some work to the next quarter and try to tackle it immediately. An alternative is to buffer up the pipe in memory and then send it across all at once. Obviously not ideal, but not too hard with the primitives we currently have.