[test] Test if implicit build works for correct average

If the correct average is provided, compress() may still
fail if more than the average numebr of entries are added
to an early row (e.g. the first one)

dune/istl/test/bcrsimplicitbuild.cc

@@ -331,6 +331,24 @@ void testImplicitMatrixBuilderExtendedConstructor()
   setMatrix(m);
 }
 
+
+void testAverageStorage()
+{
+  using M = Dune::BCRSMatrix<Dune::FieldMatrix<double,1,1>>;
+  for(auto j : {5, 0})
+  {
+    std::cout << j << std::endl;
+    M m(6, 6, 1, 0.0, M::implicit);
+
+    for(std::size_t i=0; i<6; ++i)
+      m.entry(j,i) = i;
+
+    m.compress();
+  }
+}
+
+
+
 int main()
 {
   int ret=0;
@@ -353,6 +371,7 @@ int main()
     ret+=testConstBracketOperatorBeforeCompress();
     testImplicitMatrixBuilder();
     testImplicitMatrixBuilderExtendedConstructor();
+    testAverageStorage();
   }catch(Dune::Exception& e) {
     std::cerr << e <<std::endl;
     return 1;

[Jö Fahlke]

@carsten.graeser Does this test still test the right thing? I gather from @dominic's answer in #21 (comment 19836) that this is really a documentation issue.

Otherwise, could you revert it? It makes the CI less useful.

[Carsten Gräser]

@joe : It does, because the behaviour is still inconsistent with the documentation. There are two ways to fix this: Either fix the implementation or be honest and document what it really does and not what we wish it should do. As it is, I consider this a bug that is triggered by the test. We should maybe start a general discussion on the purpose of the CI.

[Carsten Gräser]

I opened infrastructure/issues#32 in order to discuss the general question how to deal with such cases.