Testsuite: Fix globbing command for all-headers tests, cleanup code
Turns out that globbing together a full path and manually extracting
relative paths is not the best idea. Clean up this mess by using
FILE(GLOB_RECURSE [base path] [...]) that produces nice relative paths.
Merge pull request #131 from bangerth/sort-changelog
A couple of entries had ended up in the wrong place (general changes instead of specific changes). While it's debatable where exactly the line is, these items should probably be in the second category. Move them there.
* The check for support of -(f)openmp-simd is now done in
checks/check_01_cpu_features.cmake and guarded by
DEAL_II_ALLOW_PLATFORM_INTROSPECTION
The guard is important in order to allow for platform independent
compilations (as required by Linux distributions)
Furthermore, this allows for manual override via
DEAL_II_HAVE_OPENMP_SIMD.
* The pragma statement is refactored into a macro defined in
config.h.in. With this, no -Wno-unknown-pragmas workarounds are
necessary and picky compilers that choke on "#pragma" are happy.
Replace the original attempt at providing step-53 by something real.
In particular, provide a step-53 that shows how to describe geometries
using charts and use a section of Earth under Eastern Africa as an
example. This has about the right level of complexity, and produces
awesome pictures.
Replace the original attempt at providing step-53 by something real.
In particular, provide a step-53 that shows how to describe geometries
using charts and use a section of Earth under Eastern Africa as an
example. This has about the right level of complexity, and produces
awesome pictures.
This commit was generated from a number of smaller one:
More text.
Implement the Africa model for the WGS 84 geometry (actual topography still missing).
Provide the actual data set. Provide a way of reading it when not compressed.
Results and finish the program. Much still to be done.
Timo Heister [Fri, 29 Aug 2014 20:25:39 +0000 (16:25 -0400)]
Suppress clang warnings when compiling bundled boost 1.56
Clang generates a bunch of warnings of the type:
bundled/boost-1.56.0/include/boost/preprocessor/tuple/eat.hpp:23:26: warning: variadic macros are a C99 feature [-Wvariadic-macros]
bundled/boost-1.56.0/include/boost/mpl/if.hpp:131:1: warning: empty macro arguments are a C99 feature [-Wc99-extensions]
This patch just disables those warnings. Also see https://code.google.com/p/dealii/issues/detail?id=252
Timo Heister [Mon, 25 Aug 2014 15:24:23 +0000 (11:24 -0400)]
Merge pull request #118 from luca-heltai/fix_travis_ci
Removed all builds. Only indent test.
I implemented here Matthias' suggestion. Removing all unnecessary comments and lines from the Travis CI configuration scripts. This should also make it much faster, since we don't need to install anything anymore (except astyle). This is the one that closes #89.
Timo Heister [Mon, 25 Aug 2014 14:08:35 +0000 (10:08 -0400)]
Merge pull request #117 from luca-heltai/travis_ci_indent_check
Only test indentation on pull requests. This closes issue #89, where we discussed about possibly testing with Travis CI the indentation of the pull request.
Since most of the pull requests are failing anyway on Travis (time limit for a pull request is 25 minutes, and we hit it pretty regularly), this pull request configures Travis CI to only check indentation in pull requests, and do the full build only when the pull request is merged on master.
Import those parts of BOOST 1.56.0 that we need for deal.II.
This includes renaming the existing directory and overwriting every
file with the corresponding one from BOOST 1.56.0. Then removing thos
parts of the boost sublibraries that we don't need but that for
some reasons were included previously. Specifically, this applies to
documentation and build directories in libs/ and to the fusion and
spirit libraries that consist exclusively of header files.
Use an explicit cast when need a bool from optional<...>. source/base/parameter_handler.cc
The implicit cast was previously allowed, but newer versions of BOOST together with C++11 mark
the cast as 'explicit', and it fails in the current context.
Matthias Maier [Thu, 21 Aug 2014 17:10:48 +0000 (19:10 +0200)]
CMake: Remove all "help" targets
It turns out that CMAKE_PROPERTY is too buggy to be useful. We cannot
set property CMP0037 to the old behavior in order to define a custom "help"
target in a macro without a warning. o_O
Luca Heltai [Thu, 21 Aug 2014 13:28:37 +0000 (15:28 +0200)]
Merge pull request #103 from bangerth/move-functions
These functions were previously in GridTools, but conceptually they create
meshes and so should be in GridGenerator. Move them there but leave the
old functions in GridTools with a note that they are now deprecated.
Also structure the documentation of the functions in GridGenerator better.
Move create_union_triangulation() and extract_boundary_mesh().
These functions were previously in GridTools, but conceptually they create
meshes and so should be in GridGenerator. Move them there but leave the
old functions in GridTools with a note that they are now deprecated.
Also structure the documentation of the functions in GridGenerator better.
Matthias Maier [Thu, 21 Aug 2014 09:04:09 +0000 (11:04 +0200)]
Fix memory_consumption_01 for icc
The comparison file for compiler=Intel=yes was just invalid. Current icc
does not miscompile the test in release mode anymore. So remove all special
comparison files.
Matthias Maier [Thu, 21 Aug 2014 08:24:52 +0000 (10:24 +0200)]
Merge pull request #88 from guidokanschat/fe_from_name_codim_ready
FETools::get_fe_from_name() implemented for nonzero codimension
- Existing testbits/get_fe_from_name passes
- Modified all get_name functions to include the spacedim parameter. …
- Added one test for codimension one finite element generation from names.
- Reintroduce get_fe_from_name<dim>() with one template parameter as deprecated
- New function get_fe_by_name<dim,spacedim>()
Luca Heltai [Sat, 16 Aug 2014 16:47:09 +0000 (18:47 +0200)]
Modified all get_name functions to include the spacedim parameter.
Added one test for codimension one finite element generation from names.
Reintroduce get_fe_from_name<dim>() with one template parameter as deprecated
New function get_fe_by_name<dim,spacedim>()
Timo Heister [Wed, 20 Aug 2014 19:02:09 +0000 (15:02 -0400)]
Merge pull request #100 from bangerth/silence-warning-2
Silence a warning from gcc 4.9.1. The warning pertained to the fact that we define a function in an anonymous
namespace that is only called from a function inside an #ifdef ... #endif
block. The solution is to move the function into #ifdef ... #endif as well.
This addresses this test result: http://cdash.kyomu.43-1.org/viewBuildError.php?type=1&buildid=11683
I get an error of the form
/u/bangerth/p/deal.II/1/dealii/include/deal.II/meshworker/dof_info.h:69:9: runtime error: load of value 216, which is not a valid value for type 'bool'
when running the testsuite with the ASAN framework. This is caused by the DoFInfo
class having a member variable that is left uninitialized in the constructor.
Fix this.
The warning pertained to the fact that we define a function in an anonymous
namespace that is only called from a function inside an #ifdef ... #endif
block. The solution is to move the function into #ifdef ... #endif as well.
This addresses this test result: http://cdash.kyomu.43-1.org/viewBuildError.php?type=1&buildid=11683
Timo Heister [Wed, 20 Aug 2014 18:24:03 +0000 (14:24 -0400)]
Merge pull request #99 from bangerth/fix-test-signature
Rename output of tests that need PETSc, not Trilinos. The tests actually use PETSc, but their output names suggested that they need Trilinos. This was wrong.
Luca Heltai [Tue, 19 Aug 2014 13:35:24 +0000 (15:35 +0200)]
Merge pull request #94 from bangerth/silence-warning
Some newer compilers kept warning about an out-of-bounds array access
in Point::operator() but didn't indicate where exactly it came
from. This turned out to be in DataOutBase::write_eps(), a function
that is implemented only for dim==2 but was compiled also for dim==1
and dim==3 and contained this statement: 'pointsvertex'. The patch
reorganizes the code so that the function is only compiled for dim==2
and a more general specialization that does nothing but abort is used
for all other dimensions.