[Dune] first release candidate for 2.6 release

[Martin Nolte]

Hi Simon,

thanks for this excessive testing. Please note that we never officially support
the Intel compiler, because it cannot be freely use by all developers. But, as
you pointed out, we should try to make Intel 2016+ work and adapt our
"should-work" note accordingly.

As for the two issues you reported: Could you file them on GitLab and label them
as relevant for the 2.6 release?

Again, many thanks for testing.

Best,

Martin

On 01/08/2018 10:19 AM, Simon Praetorius wrote:
> Thanks for providing this release candidate,
> 
> I have tried to compile dune 2.6.0rc1 on our cluster Taurus a TU-Dresden with
> gcc and intel compilers and have encountered some problems I want to share here.
> Some are minor and some are bigger issues:
> 
> My library setups is:
> ---------------------
> CMake 3.9.0, BullXMPI 1.2.8.4, METIS 5.1.0, ParMETIS 4.0.3, MKL 2015/2017
> 
> 
> Compiler GCC 6.2.0:
> -------------------
> - (different) warnings: 7
> - errors: 1
> 
> dune-istl/dune/istl/repartition.hh:1381:27: error: cannot convert 'float*' to
> 'real_t* {aka double*}' for argument '10' to 'int ParMETIS_V3_PartKway'
> 
> The error is related to how ParMETIS is compiled on a system. On my system it
> uses `double` pointers. See appended text for a simple solution
> 
> 
> compiler Intel 2015 (15.0.3):
> -----------------------------
> - cmake: Unable to determine C++ standard support for the compiler. Needs to set
> manually DISABLE_CXX_VERSION_CHECK=1, CXX_MAX_SUPPORTED_STANDARD=14, and
> CMAKE_CXX_FLAGS="-std=c++14"
> - warnings: > 3
> - errors: > 8 (some are related errors)
> 
> main issue: intel 15 is compatible with gcc up to 4.9 (and uses libstdc++ and
> headers from this gcc version on my system)
> -> e.g. std::is_trivially_copyable not implemented
> -> no generic lambda (or at least not fully supported)
> -> constexpr functions can not return void
>    (
> https://stackoverflow.com/questions/29261276/constexpr-void-function-rejected )
> 
> 
> compilers intel 2016-2017:
> --------------------------
> similar as above. Number of errors reduced by not to 0
> 
> 
> compiler intel 2018 (18.0.1):
> -----------------------------
> - warnings: 7
> - errors: 6
> 
> 
> The errors are all of the same kind on different places in the code and related
> to, what I think, is a bug in the intel compilers. In a Hybrid::forEach, inside
> the lambda, the names from namespace Dune::Hybrid are not found, although `using
> namespace Dune::Hybrid` is stated before the call to `forEach`. These errors are
> simple to workaround, i.e. simply add an additional `using namespace
> Dune::Hybrid` inside the lambda. But this is not very beautiful and will be
> forgotten probably in implementations in the future. Maybe we have to formulate
> a style-guide for `Hybrid::forEach` calls.
> 
> 
> --------
> 
> Since, intel 2015 is limited to gcc 4.9, I think the support may be dropped. But
> intel 2016 supports gcc 5.1 and thus I hope, we could make Dune compatible with
> this compiler at least, maybe with some small workarounds.
> 
> See appended the list of warnings and errors I've found when just compiling the
> dune modules and tests from dune-common, dune-geometry, dune-grid, dune-istl,
> dune-localfunctions.
> 
> Best wishes,
> Simon Praetorius
> 
> 
> Zitat von Andreas Dedner <a.s.dedner at warwick.ac.uk>:
> 
>> Hi everyone.
>>
>> We just wanted to let you know that just in time for Christmas the first
>> release candidate of the 2.6 release is available for testing.
>> Tar balls are available
>> https://www.dune-project.org/releases/2.6.0rc1/
>> or use the v2.6.0rc1 tag for helping us, by testing the release with
>> your own modules, system, workflow... Let us know if you encounter any
>> problems. Also please skim through the change log (same link as above)
>> to see if something is unclear or missing.
>>
>> Wish you all a merry Christmas and a happy new year
>> Andreas and Martin
>>
>> _______________________________________________
>> Dune mailing list
>> Dune at lists.dune-project.org
>> http://lists.dune-project.org/mailman/listinfo/dune
> 
> 
> 
> 
> 
> _______________________________________________
> Dune mailing list
> Dune at lists.dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune
> 

-- 
Dr. Martin Nolte <nolte at mathematik.uni-freiburg.de>

Universität Freiburg                                   phone: +49-761-203-5630
Abteilung für angewandte Mathematik                    fax:   +49-761-203-5632
Hermann-Herder-Straße 10
79104 Freiburg, Germany