BoundingBoxTree.py tests fails with GCC 4.8

[Prof Garth Wells]

FAIL: test_compute_closest_entity_3d (__main__.BoundingBoxTreeTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "BoundingBoxTree.py", line 372, in test_compute_closest_entity_3d
    self.assertEqual(entity, reference[0])
AssertionError: 4 != 2

This problem causes a bunch of demos to fail too.

[Anders Logg (Chalmers)]

Does it work now? (fix pushed to next)

[Prof Garth Wells]

No. The demos that use the geometry code are failing, e.g.

demo/documented/auto-adaptive-poisson

[Anders Logg (Chalmers)]

In what way? I don't have gcc 4.8. I assume it's in Ubuntu 13.10 which I haven't upgraded to yet.

[Prof Garth Wells]

Here's one (demo/undocumented/auto-adaptive-navier-stokes/cpp) of a considerable number of failures:

terminate called after throwing an instance of 'std::runtime_error'
  what():  

*** -------------------------------------------------------------------------
*** DOLFIN encountered an error. If you are not able to resolve this issue
*** using the information listed below, you can ask for help at
***
***     [e-mail]
***
*** Remember to include the error message listed below and, if possible,
*** include a *minimal* running example to reproduce the error.
***
*** -------------------------------------------------------------------------
*** Error:   Unable to complete call to function compute_closest_entity().
*** Reason:  Assertion closest_entity < std::numeric_limits<unsigned int>::max() failed.
*** Where:   This error was encountered inside /home/garth/code/fenics/dev/dolfin.d/dolfin-master/dolfin/geometry/GenericBoundingBoxTree.cpp (line 235).
*** Process: 0
*** -------------------------------------------------------------------------

[Anders Logg (Chalmers)]

Strange. Update to 13.10 in progress. I'll take a closer look when I have gcc 4.8.

[Marco Morandini]

For what it's worth it works for me, but I use gcc 4.8.1 (and opensuse).

[Anders Logg (Chalmers)]

After spending 8 hours tracking down this bug, I finally found it. The bug is in std::nth_element, more specifically in /usr/include/c++/4.8/bits/stl_algo.h.

So not my fault this time! :-)

The bug is fixed in gcc 4.8.3. I don't know when it will enter Ubuntu, but until then the fix is to edit the file above and replace -2 by -1 in lines 2282 and 2294.

[Prof Garth Wells]

Well found!!

Here's the GCC bug report:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58800

[Anders Logg (Chalmers)]

Not a bug in DOLFIN - bug in the C++ standard library which has been fixed upstream.

[Prof Garth Wells]

Ubuntu bug report: https://bugs.launchpad.net/bugs/1246802