dash::Array storing the wrong team?

[devreal]

I am still debugging #292 (with increasing desperation). I found the following strange behavior

TEST_F(ArrayTest, TeamSplit)
{
  auto & team_all = dash::Team::All();
  auto   ext_x    = team_all.size();

  if(team_all.size() < 2){
    SKIP_TEST();
  }
  if(!team_all.is_leaf()){
    LOG_MESSAGE("team is already splitted. Skip test");
    SKIP_TEST();
  }

  dash::barrier();

  LOG_MESSAGE("Splitting team ...");
  auto & myteam = team_all.split(2);
  LOG_MESSAGE("... Team split finished");

  dash::Array<double> array_a(ext_x, myteam);
  ASSERT_EQ_U(array_a.team(), myteam); // << this test fails!

  array_a.barrier();
  // Check if array is allocated
  ASSERT_NE_U(array_a.lbegin(), nullptr);
  team_all.barrier();
}

I added the check ASSERT_EQ_U(array_a.team(), myteam), which surprisingly fails. My understanding is that array.team() should yield the team the array was created with so the check should succeed. Looking through the code, I don't see what could go wrong there. Ideas?

UPDATE: The error I get is

[  ERROR   ] [UNIT 0] in /home/joseph/src/dash/dash/dash/test/container/ArrayTest.cc:225
      Expected: array_a.team()
      Which is: dash::Team(0)
To be equal to: myteam
      Which is: dash::Team(1.0)

[ddiefenthaler]

I think I found the cause of this strange behaviour,
If I'm right it is related to PR #304 and through the implicit conversion from dash::Team to dash::TeamSpec now the wrong contructor gets called during the construction of dash::Array (Code).

The fix for this should be simple.

[ddiefenthaler]

Acctually it is not so simple, as the constructors of the different pattern types mismatch now.
For example the CSRPattern has no constructor which takes a TeamSpec.

@fuchsto Your feeling about the missing explicit was quite appropriate.

My best bet to fixing this issue is to revert PR #304 and just remove the default parameter definitions instead of the whole functions...
By the way who added those default parameters in the first hand? The TeamSpec should be dependend on the distribution which is not possible this way, if I got this right. For at least in line 336 of the deleted function it looks that way: diff #304