I'm attaching the stack when the exception is generated. The problem seems to be that nsJSUtils::ReportPendingException is entering some sandbox scope when trying to report the exception. Consequently, the exception is wrapped in a filtering wrapper. Could this be a problem similar to bug 1066340? Does nsFrameScriptExecutor::LoadFrameScriptInternal need to use AutoEntryScript?

Yes, that should fix this bug. Add a patch and flag me for review? Jimb - note that more of the AutoEntryScripts are being added, which will need fixing up in your patch in bug 971673.

Are you sure? I tried this patch and it doesn't work. Looking more closely, nsJSUtils::ReportPendingException is calling GetScriptContextFromJSContext, which is here: http://mxr.mozilla.org/mozilla-central/source/dom/base/nsDOMJSUtils.h#16 That code is checking if the JSContext private is an nsISupports. Looking at JS_SetContextPrivate, it looks like we only set that for nsJSContext. And I'm guessing we just use the safe JS context here. So isn't this code bound to fail?

Oh I see. I'd been assuming that the JSObject we'd be using to init the AutoEntryScript was associated with a DOM window, but I guess that's not the case with frame scripts. Given that, the problem is more that nsJSUtils::ReportPendingException is broken. So we need to fix bug 981187 to make progress here. Setting the dep.

Is there any way we can fix this sooner? Bug 981187 sounds like a pretty big task. I'd really like to have a fix for this in the next few days. In https://bugzilla.mozilla.org/show_bug.cgi?id=1048164#c14 the author of Greasemonkey is trying to convert his code to e10s using content scripts and is getting this message. If add-on authors can't convert their code, that's a really serious problem.

The current error reporting machinery is basically broken. There's not a great way to spot fix it. I've been hacking on this, and I think I can get us to where we need to be with a relatively defined amount of work. Give me a day or two.

I think we might actually be able to spot-fix this after all. Let me investigate.

Oh, looks like there isn't actually a testcase in this bug. Bill, what happens if you take your existing patch and also replace the call to nsJSUtils::ReportPendingException with a direct call to JS_ReportPendingException?

The original testcase I had for this was something weird, and now I don't have it anymore. I haven't had much luck writing a new testcase.

Ok. Bug 1070842 is also not far from landing anyway.

Bill, can you verify that any issues here are now resolved?

Yeah, let's close this. I'll file something if anything else pops up. Thanks.