[Trilinos-Users] Tpetra example compilation error

[Martin Vymazal]

Dear Trilinos developers,

 the example 

 "Create and use a preconditioner for a sequence of linear systems"  

from 

  https://code.google.com/p/trilinos/wiki/Tpetra_Belos_CreateSolver 

does not compile with local_ordinal_type set to unsigned integer. 

When I typedef

typedef unsigned local_ordinal_type;

/home/martin/local/trilinos-11.2.3/lib/cmake/Trilinos/../../../include/BelosTsqrOrthoManagerImpl.hpp:1545:7: 
error: no matching function for call to 
‘Tpetra::TsqrAdaptor<Tpetra::MultiVector<double, unsigned int, long int, 
Kokkos::TPINode> >::factorExplicit(Tpetra::MultiVector<double, unsigned int, 
long int, Kokkos::TPINode>&, Tpetra::MultiVector<double, unsigned int, long 
int, Kokkos::TPINode>&, Teuchos::SerialDenseMatrix<int, double>&, bool&)’
       tsqrAdaptor_.factorExplicit (X, Q, B, forceNonnegativeDiagonal_);
       ^

I don't know if that's a bug or feature, but I suppose this should normally 
compile with unsigned integers used as local ordinal type.

Could you explain why is (signed) integer type used for indexing entries in 
matrices and vectors? I understand that subtracting two unsigned integers can 
give unexpected results when not handled with care; but besides this reason, 
why are signed integers preferred?

Best regards,

 Martin Vymazal