Zarro Leaks found.

Well, almost.

The "open browser and close again" as well as the "open mail and close again" tests our (Linux) bloat test box is doing both reports 0 refcnt leaks now, even without the slight hack we had in there until today (not loading the EM RDF datasource and not populate the "View > Apply Theme" menu with actual themes).

And even the chrome mochitest suite reports "no leaks" on Linux and Windows.

Unfortunately, we somehow still leak 1112 bytes on Mac for those tests. For the plain mochitest suite, where we run much more tests, we report 8 bytes leaked on Windows and Linux, on Mac OS, we report those 8 plus the previously mentioned 1112, making up 1120 bytes total. And for the browser chrome mochitests, we leak 1112 bytes on Linux and Windows, and 1232 on Mac. Somehow this 1112 number is something we like - interestingly, we get it in those places where one non-browser window is open in addition to the standard browser window (the "hidden window" on Mac or the additional window for browser chrome tests), I wonder if that's somehow connected. Also, we're leaking the RDFServiceImpl and an InMemoryDataSource there, which could be an indication what to look for, I guess (unfortunately I don't know the code too well myself).

By the way, I just worked a long-standing workaround to our EM RDF datasource leak into an actual patch, removing the datasource in the unload handler of the browser window, the real fix would be in cycle collection though.

Oh, and have patches waiting for review to get the actual test failures on SeaMonkey machines out of the way and hopefully turn them green for the holidays.