MultiTypeBlockVector: Inherit tuple constructors

• Makes member function count const
• Inherits tuple constructors so that MultiTypeBlockVector can be constructed from existing Vectors, or we can even construct it with references to existing vectors.
• Test construction from references in the test
• Fixes the static_assert in the test that shouldn't really have compiled before

Fails for clang with

/builds/tkoch/dune-istl/dune/istl/test/multitypeblockvectortest.cc:32:19: error: static_assert expression is not an integral constant expression
    static_assert(multiVector.size()==2, "Method MultiTypeBlockVector::size() returned wrong value!");
                  ^~~~~~~~~~~~~~~~~~~~~
/builds/tkoch/dune-istl/dune/istl/test/multitypeblockvectortest.cc:90:3: note: in instantiation of function template specialization 'testMultiVector<Dune::BlockVector<Dune::FieldVector<double, 3>, std::allocator<Dune::FieldVector<double, 3> > >, Dune::BlockVector<Dune::FieldVector<double, 1>, std::allocator<Dune::FieldVector<double, 1> > > >' requested here
  testMultiVector(multiVector);
  ^

However I don't see how this is related to the changes in this branch. I guess because multiVector is now tested in the function scope somehow?

added 1 commit

• b427926f - [test][multitypeblockvector] Test construction of MultiTypeBlockVector of references of Vectors

Compare with previous version

@oliver.sander can you comment on this proposal? Is there a reason MultiTypeBlockVector only has a default constructor?

(... I don't see why the old static_assert worked. The LHS of the . is always evaluated, even if the RHS denotes a static member. I guess the compiler was a bit permissive ...)

added 1 commit

• d045b48c - [test][multitypeblockvector] Test construction of MultiTypeBlockVector of references of Vectors

Compare with previous version

There was never a particular reason to only have the default constructor. I do not object to your changes.

added 1 commit

• 6020d45d - [test][multitypeblockvector] Test construction of MultiTypeBlockVector of references of Vectors

Compare with previous version

changed the description

ok cool. I would like to use this feature in our upcoming dumux release so I would ask for a backport to 2.6 too once this is merged.

(... OK, new hypothesis: the old static_assert worked because clang never does the lvalue-to-rvalue conversion for the lhs of a .. If the lhs is a function-local automatic variable, the "identity" (address) of that variable is known sufficiently to be constexpr, even if the value is not. If the lhs is passed into the function as a reference, then it's identity is definitely not constexpr within that function. Of course, to add to the confusion, only clang seems to be strict about that, gcc considers the result constexpr either way...)

@tkoch , I will merge this, but please add a short note to CHANGELOG.md first.

added 1 commit

• ed7eaecc - [changelog] Add entry for new MultiTypeBlockVector constructors

Compare with previous version

added 1 commit

• ce4a4d94 - [changelog] Add entry for new MultiTypeBlockVector constructors

Compare with previous version

enabled an automatic merge when the pipeline for ce4a4d94 succeeds

merged

mentioned in commit 54ee6051

mentioned in commit tkoch/dune-istl@448c0881

mentioned in merge request !242 (merged)

mentioned in commit bb0fee97

mentioned in merge request !395 (merged)