I noticed with my patches in bug 59314 that I was triggering an assert in nsJSContext::LeaveModalState()... "Uh, mismatched LeaveModalState() call!". A bit of debugging revealed that nsGlobalWindow::LeaveModalState was getting a different context that the one it got in ::EnterModalState. bkap explains that, basically, in the EnterModalState call it's picking up the existing content (chrome?) context. But since the 2nd event loop is still on the stack when calling LeaveModalState, there's a null context pushed, and thus the obtained context is that of the JS component where this code lives. After a bit of discussion with jst, there was conclusion that we should (1) use the window's context to avoid this problem and (2) file a followup bug to improve how we track modal state. This patch is still compiling. ;)

...and filed bug 611170 for improving the state tracking.

Comment on attachment 489673 [details] [diff] [review] Patch v.1 a=me since this is simple and blocks a blocker.

Pushed http://hg.mozilla.org/mozilla-central/rev/de3b5061113f