On 10/18/2012 09:54 AM, Brendan Eich wrote: > > Terrence Cole wrote: >> On 10/17/2012 10:01 PM, Nicholas Nethercote wrote: >>> Terrence and I have had problems with that one before. IIRC the test >>> has some bogosity about it, but I don't remember the details -- >>> Terrence, do you? >> >> I'll just echo what Nicolas said: anything that changes the interpreter >> stack will randomly make this test explode in this way. > > That test is not just some old dog to take out back and shoot. If it > barks, others will too. It's also testing things that are being > standardized in ES6.
I'm sorry, I didn't mean to imply that that was what we should do: it's clearly our problem and not the test's -- although I'll freely admit that after the first few instances I read over that test with every intention of going ol' yeller on it. > We have a zero regression policy, project-wide (with short waivers > clearing orange by patching forward instead of backing out. For > SpiderMonkey we need either to fix a regression or remove a test that > was counting on a bug or misfeature that's not being standardized. This test doesn't appear to be orange on any of the toolchains used by tinderbox at the moment, otherwise the regressing change would have been backed out by the sheriff. > So in this case, we need a fix -- is there a bug on file? I haven't seen one this week. I'm able to repro against tip with dbg-clang-x64 so I'll file and look into it today. >> That said, I >> also hit this a bunch when implementing the StoreBuffer and that code >> explicitly goes nowhere near the stack. I vaguely remember that someone >> looked into it and found their compiler randomly deciding to give >> js::Interpret a bogusly humongous frame. > > These bugs come up from time to time. I just noticed > > https://bugzilla.mozilla.org/show_bug.cgi?id=799494 Thanks for the link. > still open. > >> I was /really/ hoping that unraveling the threaded interpreter would >> make this problem go away permanently. That work landed over a week ago >> in 109802:44079242ee9b. Brendan, what revision are you testing? > > beich-10880:src brendaneich1$ hg tip > changeset: 110531:366998e4e1dc > tag: tip > user: Jonathan Kew <[email protected]> > date: Wed Oct 17 20:50:06 2012 +0100 > summary: bug 794038 followup - add #include "nsIPresShell.h" to > nsChildView.mm to avoid relying on fragile indirect inclusion. rs=Ms2ger > > /be Oh well, it was worth a shot. -Terrence _______________________________________________ dev-tech-js-engine-internals mailing list [email protected] https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
