Some things to consider: 1. On windows, breakpad used to be wired in test_shell. And I'm pretty sure we used to archive crash dumps for the layout tests too. It should not be hard to do that again. Huan also write a nice script to dump to stdio the crashing stacks of all crashes that happened in a buildbot run. It only runs on "chromium xp" for now, but there is no reason why we could not make it work for the layout tests. See http://build.chromium.org/buildbot/waterfall/builders/Chromium%20XP/builds/6491/steps/Process%20Dumps/logs/stdio for example.
2. The show stopper for any implementation of this feature is that the machines running the layout tests don't have the pdbs for test_shell. Since the binary is built on another machine, it was too slow to copy the pdbs from one machine to another. If you guys think it's important, and can take the ~30-60 more seconds to cycle the bots, then we can copy them too, and the feature would work. Nicolas On Sat, Aug 8, 2009 at 11:42 AM, Albert J. Wong (王重傑) <ajw...@chromium.org>wrote: > FYI, the code in debug_util.h will generate a stack trace, but symbol > resolution doesn't work on mac. Last I messed with it (~4 months ago), mac > didn't work because most of the symbols are private. Mark Mentovai > suggested trying to reimplement dladdr, but I could never get it working. > Here's the uploaded code if anyone to mess with it: > > http://codereview.chromium.org/164228 > > On windows and linux, assuming you actually have symbols generated (which I > don't think you do for windows on the build bots), getting a trace should be > as simple as creating one of those StackTrace objects in debug_util.h, and > calling PrintBacktrace or OutputToStream on it. The hard part is knowing > when to create the object, and making sure you're on the right thread. > Also, these functions do some heap allocations so using them in a crash > handler might be a bit unsafe...but if it's crashing, and there's already no > way to get a core or something, making it crash harder isn't going to > matter. > > -Albert > > > > On Sat, Aug 8, 2009 at 11:12 AM, Jeremy Orlow <jor...@chromium.org> wrote: > >> I'll take a look. >> If anyone has ideas on how to implement this (beyond looking at >> base/debug_util.h) please let me know! The last time I messed around with >> programatic stack traces was 5+ years ago. :-) >> >> >> On Sat, Aug 8, 2009 at 7:53 AM, Dimitri Glazkov <dglaz...@chromium.org>wrote: >> >>> Somebody please run with this! :) >>> >>> :DG< >>> >>> On Fri, Aug 7, 2009 at 8:45 PM, Darin Fisher<da...@chromium.org> wrote: >>> > On Fri, Aug 7, 2009 at 8:39 PM, Ojan Vafai <o...@chromium.org> wrote: >>> >> >>> >> On Fri, Aug 7, 2009 at 8:45 PM, Jeremy Orlow <jor...@chromium.org> >>> wrote: >>> >>> >>> >>> Has anyone ever looked into printing out stack traces when layout >>> tests >>> >>> crash? Of the couple layout test crashes I've investigated, I think >>> most >>> >>> could have been solved just by having a stack trace. I'm not really >>> sure >>> >>> what's involved or if anyone's looked into this, which is why I'm >>> asking. >>> >>> This could be especially helpful for flaky tests that crash. >>> >> >>> >> I don't remember anyone having looked into this. I agree that this >>> would >>> >> make tracking down and fixing these issues immensely easier, >>> especially for >>> >> tests that only sometimes crash. >>> >> Ojan >>> > >>> > I've wanted this for a very long time. I think there might be a bug on >>> it. >>> > At any rate, we now have a nice utility in base/debug_util.h that can >>> > provide a stack trace. I would love to see crash stacks on the >>> buildbot. >>> > It would probably help us eliminate a lot of flakiness! >>> > -Darin >>> > > >>> > >>> >> >> >> >> > > > > --~--~---------~--~----~------------~-------~--~----~ Chromium Developers mailing list: chromium-dev@googlegroups.com View archives, change email options, or unsubscribe: http://groups.google.com/group/chromium-dev -~----------~----~----~----~------~----~------~--~---
