I think this explains why sometimes I can't connect to B2G. I need to restart B2G to fix this issue. Not clear STR yet.

Most likely because of this loadFrameScript, evaluating child.js multiple times in the app frame: http://mxr.mozilla.org/mozilla-central/source/toolkit/devtools/server/actors/webapps.js#796

Comment on attachment 8364250 [details] [diff] [review] loadFrameScript, execute scripts only once Currently, each time you call loadFrameScript, the related script is executed a new time. We discussed with Vivien and tend to agree on the fact it should be executed only once. Do you think we could enforce such behavior in loadFrameScript? Or it is up to the caller to ensure calling it once? If you think that's ok, I'll write some test and push it to try, otherwise I'll address that in the call site.

It is up to the caller. Using data: url is very useful for testing and I certainly want those to be executed every time I pass such to loadFrameScript.

Ok, so in this patch, we just let the frame script being evaluated more than once and prevent any exception due to const or strict mode.

The patch looks big, but that's just a indentation shift, I'm encapsulation the script in an anonymous function.

Does this change mean we'll have an extra set of actors running each time you start debugging an app? As in, if I start debugging app 1, close the toolbox, then start debugging app 1 again on the same connection, are there now 2 sets of actors loaded for this app? I'm wondering because it doesn't seem like we clean up all the things being added here.

Hum in that particular scenario, the appActorsMap in the webapps actor will prevent spawing another actor, but that's a still a relevant question for the current case. We will spawn another actor whereas the one spawned during the previous connection is still around. They will all be cleaned up once the app is killed. There is multiple improvement to be made: - destroy all actors on disconnect? We never do such thing yet. We never try to clean up stuff on the child side. It is not really clear what we should be doing to clean stuff properly on child side. Would calling `disconnect()` on the ContentAppActor (which is a BrowserTabActor) be enough?? - factor out connectToApp (bug 962577) and may be also the actor map. Move it to a global component that would keep the actor alive between connections.

The current behavior in all other cases is that when the connection is gone, all actors will be GCed. I think the child process actors should behave the same.

We will improve actors cleanup in a followup as this breakage is quite painfull and current patch doesn't improve/regress actors cleanup.

See bug 966991 for followup.