patch: break into debugger on death in sanitizers

[Reid Kleckner]

I bet it would work fine, but the main thing is to do things in such a way
that gdb works out of the box without having to write documentation or give
workarounds.  Printing a stack trace would work better.

The sanitizers use a raw exit syscall to quit, rather than calling _exit
from libc, and this bypasses gdb's breakpoints.  This is probably a gdb
bug, but it'd be nice if the program stopped in the debugger before exiting.


On Tue, Oct 29, 2013 at 1:49 PM, Kostya Serebryany <kcc at google.com> wrote:

> btw, why is the currently existing flag abort_on_error=1 not enough for
> gdb?
>
>
> On Tue, Oct 29, 2013 at 1:26 PM, Reid Kleckner <rnk at google.com> wrote:
>
>> Based on in person discussion, this is not an option, because there are
>> users of ptrace that aren't debuggers.  See lsan and heapchecker for
>> example.
>>
>>
>> On Sat, Oct 26, 2013 at 3:50 AM, Alexander Potapenko <glider at google.com>wrote:
>>
>>> It should be easy to detect whether we're running under gdb (e.g. if
>>> we're being ptraced; perhaps something needs to be done in the presence of
>>> seccomp-bpf)
>>> On Oct 25, 2013 10:55 PM, "Kostya Serebryany" <kcc at google.com> wrote:
>>>
>>> I am not opposed to this change, but I'd like to test it a bit more
>>> carefully to make sure none of our users get angry on us for behavior
>>> change.
>>> I'd suggest to put this sigtrap under a flag (off by default) and then
>>> flip the default to true after a bit of testing.
>>> We already have abort_on_error, may add sigtrap_on_error or some such.
>>>
>>> (may not reply until Mon next time)
>>>
>>>
>>> On Fri, Oct 25, 2013 at 7:26 PM, Chandler Carruth <chandlerc at google.com>wrote:
>>>
>>>> Sigtrap, which seems less likely to cause deep surprise in the presence
>>>> of signal handlers.
>>>>  On Oct 25, 2013 3:56 AM, "Kostya Serebryany" <kcc at google.com> wrote:
>>>>
>>>>>  So this will change the default behavior from doing exit(1) to doing
>>>>> abort(), right?
>>>>>
>>>>>
>>>>> On Fri, Oct 25, 2013 at 3:16 AM, Nick Lewycky <nlewycky at google.com>wrote:
>>>>>
>>>>>> Running sanitized code under gdb can be annoying because the program
>>>>>> will print out its error then exit, instead of the usual behaviour when we
>>>>>> get an abort() of stopping the program there and allowing the user to
>>>>>> inspect its stack trace.
>>>>>>
>>>>>> Add a trap to the exit path in sanitizer_common. Please review!
>>>>>>
>>>>>> Nick
>>>>>>
>>>>>> _______________________________________________
>>>>>> cfe-commits mailing list
>>>>>> cfe-commits at cs.uiuc.edu
>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> cfe-commits mailing list
>>>>> cfe-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>>>
>>>>>
>>>
>>> _______________________________________________
>>> cfe-commits mailing list
>>> cfe-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>
>>>
>>> _______________________________________________
>>> cfe-commits mailing list
>>> cfe-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131029/7b533f2d/attachment.html>