MueLu: Potential typing mistake in BelosXpetraStatusTestGenResSubNorm (edited)

[kliegeois]

Bug Report

@trilinos/muelu
@tawiesn

Description

There is, from time to time, a seg fault in file BelosXpetraStatusTestGenResSubNorm.hpp in the block of lines 289-306:

    for (; p2<curLSIdx_.end(); ++p2) {
      // Check if this index is valid
      if (*p != -1) {
        // Check if any of the residuals are larger than the tolerance.
        if (testvector_[ *p2 ] > tolerance_) {
          // do nothing.
        } else if (testvector_[ *p2 ] == Teuchos::ScalarTraits<Scalar>::magnitude(Teuchos::ScalarTraits<Scalar>::zero())) {
          reset();
        } else if (testvector_[ *p2 ] <= tolerance_) {
          ind_[have] = *p2;
          have++;
        } else {
          // Throw an std::exception if a NaN is found.
          status_ = Failed;
          TEUCHOS_TEST_FOR_EXCEPTION(true,StatusTestError,"StatusTestGenResSubNorm::checkStatus(): NaN has been detected.");
        }
      }
    }

It seems that the seg fault is due to testvector_[ *p2 ] and that *p2 is equal to -1.

I think that the test if (*p != -1) should be replaced by if (*p2 != -1) as BelosXpetraStatusTestGenResSubNorm.hpp is a template specialization of BelosStatusTestGenResSubNorm.hpp where the test is indeed if (*p2 != -1).

[kliegeois]

My bad, the seg fault was coming from another instruction.
Nevertheless, I think that if (*p != -1) is potentially a typing mistake.

[kliegeois]

Resolved by #6834 .