Re: [Libmesh-devel] Issue with negative jacobian

[David K. <dav...@ak...>]

On Tue, May 5, 2015 at 3:07 PM, David Knezevic <dav...@ak...>
wrote:

> On Tue, May 5, 2015 at 2:30 PM, John Peterson <jwp...@gm...>
> wrote:
>
>>
>>
>> On Tue, May 5, 2015 at 12:20 PM, David Knezevic <
>> dav...@ak...> wrote:
>>
>>> On Tue, May 5, 2015 at 10:27 AM, John Peterson <jwp...@gm...>
>>> wrote:
>>>
>>>>
>>>>
>>>> On Tue, May 5, 2015 at 7:10 AM, David Knezevic <
>>>> dav...@ak...> wrote:
>>>>
>>>>> [0] /home/dknez/software/libmesh-src/src/fe/fe_map.C, line 618,
>>>>> compiled May  5 2015 at 08:08:28
>>>>> ERROR: negative Jacobian: -1.79627e-07 in element 7051
>>>>>
>>>>
>>>> Is your fe_map.C not the version in master?  Line 618 doesn't have a
>>>> libmesh_error_msg() on it for me.
>>>>
>>>> In any event, we have the Elem* available, so we should probably just
>>>> be helpful and print the stuff that Roy requested before erroring out...
>>>>
>>>
>>>
>>> I added some print statements to the test code I was running before
>>> (with libMesh git hash 1a6a549f26219f462ac4ceef87c3c03e9eafa836) and I get
>>> the info printed below for the failing element.
>>>
>>> When I update to the git HEAD, I now get a segfault with no stack trace
>>> when I just loop over the elements and call fe->reinit(elem). I'm wondering
>>> if maybe ExodusIO isn't reading this mesh in properly, or something.
>>>
>>
>> That, or a dirty build?
>>
>
>
> I agree that a dirty build seems like a reasonable explanation, but it
> doesn't seem to be the case here. I did distclean, bootstrapped and
> re-configured, and I tested a bunch of libMesh examples and they all worked
> fine. But with this negative-jacobian test, l still got the same segfault.
>

OK, so the issue is that your change from today triggers an infinite loop
for my "bad element", eventually leading to the segfault.

I think the problem is that calling "elem->print_info()" then calls
"elem->volume()", which reinits an FE object, which triggers negative
jacobian error again, and then you keep looping forever.

If I comment out the elem->print_info calls that you added, I get back to
the negative jacobian error that I had before.