by testing whether we work on a end iterator. In this case the distance
function will fail. Found that out by gcc's stdlib with debugging turned on.

When using NO_DEFAULT_PATH the lib and include subdirectories will
not be searched by default by find_path and pt-scotch will not be
found outside of the standard path. This commit adds them and now
locally installed versions may supersede the system ones.

Without it linking will fail when using static libraries due
to missing symbols.

These are the *.la files below dune and the primary module library
lib<module name without dashes>.la.

This reverts commit 21dbaefe.
Thus if libtool libs are found, then we abort again.
Please do not merge to the release.

Since commit 495f2fce USE_MPI was removed from dunecontrol and
the default configuration now defaults to use
parallelism. Unfortunately, there are downstream modules that try to
support multiple versions of DUNE. Sometimes these need to decide
whether to support parallelism before they know which version of DUNE
they are using. This is was not possible when providing
--enable-parallel as no variable gets set for DUNE>=2.4.

With this commit dunecontrol always sets
CMAKE_DISABLE_FIND_PACKAGE_MPI to an appropriate value. Thus if
dunecontrol is used then one can check in one's own module whether the
variable is set explicitly to false and conclude that one is using
>=2.4 and wants parallelism. Otherwise one can check USE_MPI to detect
which behaviour is needed for 2.3.

E.g. like this:
<code>
option (USE_MPI "Use Message Passing Interface for parallel computing" OFF)

if( NOT CMAKE_DISABLE_FIND_PACKAGE_MPI STREQUAL ""
    AND NOT CMAKE_DISABLE_FIND_PACKAGE_MPI)
  set(USE_MPI ON)
endif()

if (NOT USE_MPI)
      set (CMAKE_DISABLE_FIND_PACKAGE_MPI TRUE)
 set (CMAKE_DISABLE_FIND_PACKAGE_MPI TRUE)
endif (NOT USE_MPI)
</code>

Of course without dunecontrol and using the defaults one will still
configure DUNE as parallel, but one's own module as sequential.

Actually, having these files lying around does not hurt us.
In addition there are downstream modules that especially create
these for compatibility reasons with autotools. For these
reconfiguring a module using dunecontrol currently aborts with an error

Therefore this commit resorts to just issuing a warning and proceeding
with the compilation instead of aborting with an error.

... and don't take absolute value of comparison result.
Thanks to Max Kahnt for reporting this.

The add_doxygen_target macro now takes optional parameters:
TARGET [name] : overwrite the target appendix, default is the module name
DEPENDS [list] : add a list of further dependencies for doxygen.
                               This list might include mainpage.txt
OUTPUT [name] : change the name of the output target; necessary if you don't generate html.

Otherwise it won't appear in the doxygen documentation

We follow the advice of upstream and use only empty endif() and
else() statements in new code. The removal of the optional argument
is only in generated scripts.

Add missing files, remove ancient buildsystem stuff and stuff where it is
unclear where it comes from.

[test][typetraits][warnings] Split the test for deprecated functionality in typetraits.hh into its own testcase.  That way I can simply disable deprecation warnings there.

There was a mechanism with apparently similar goal involving the define
DISABLE_DEPRECATED_METHOD_CHECKS.  However, I could not figure out how that
was intended to work, the only occurances of DISABLE_DEPRECATED_METHOD_CHECKS
were in typetraitstest.hh.  Probably you were supposed to set that define by
hand in your CPPFLAGS, which I consider unacceptable.

The section is empty since the code was split into the dune-geometry
module.

Since fda5a9e0 AlignmentOf always uses
std::alignment_of and the fallback implementation is no longer used.

It's more CMake style.

Since de430584 HAVE_DUNE_HASH is
always defined.

Nicolas Schwenck authored 9 years ago and Christoph Grüninger committed 9 years ago

floating comparisons

Some downstream modules like pdelab and dumux already abandoned the
current bahaviour of building tests on make test. The testing infrastructure
does throw them error messages during make test though. This commi introduces
the switch DUNE_DISABLE_TEST_MAGIC which may be set by downstream modules
to disable the magic. To see how to switch a downstream module, see:

http://conan2.iwr.uni-heidelberg.de/git/pdelab/dune-pdelab/commit/dc5ef4b39f5be5690457f0bbfe008b1f45d5fd98

Point to proposal for C++17 make_array(), and point out the differences of our
implementation.