AssemblerBase::reset_sparsity is nowhere used

[Jan Blechta]

AssemblerBase::reset_sparsity parameter is nowhere used (in a contradiction with documentation). Actual criterion for building sparsity pattern in AssemblerBase::init_global_tensor is

 58   if (A.size(0) == 0)
 59   {
 60     Timer t0("Build sparsity");
...
133   }
134   else
135   {
136     // If tensor is not reset, check that dimensions are correct
...
146   }

[Prof Garth Wells]

We need to keep this for a bit to avoid breaking code. It would be good if a warning could be issued when it's used. It's not a function, so it'll take something a bit cleverer to print a warning.

[Jan Blechta]

Actually, is the criterion A.size(0) == 0 intended? Is the intention that user must initialize new tensor to assemble something with different pattern?

[Prof Garth Wells]

It will soon not be possible to re-initialise a matrix/vector. This was discussed previously on the mailing list. A.size(0) == 0 is true if the matrix has not been initialised.

[Jan Blechta]

@garth-wells According to the GenericTensor interface, proper check should be A.empty() instead of A.size(0) == 0. (Use case: user (me) implements GenericMatrix and non-zero size(0) is known from construction time but tensor still needs initialization...)

Bug report: at least PETScMatrix::empty(), EpetraMatrix::empty() and uBLASMatrix::empty() seems to not work:

from dolfin import *

mesh = UnitSquareMesh(3, 3)
V = FunctionSpace(mesh, 'CG', 1)
u, v = TrialFunction(V), TestFunction(V)
a = inner(grad(u), grad(v))*dx
L = v*dx

for A in [uBLASDenseMatrix(), uBLASSparseMatrix(), PETScMatrix(), EpetraMatrix()]:
    print A.str(False), A.empty(), A.size(0)
    AssemblerBase().init_global_tensor(A, Form(a))
    print A.str(False), A.empty(), A.size(0)

for b in [uBLASVector(), PETScVector(), EpetraVector()]:
    print b.str(False), b.empty(), b.size(0)
    AssemblerBase().init_global_tensor(b, Form(L))
    print b.str(False), b.empty(), b.size(0)

[Prof Garth Wells]

@blechta 'GenericTensor::empty' makes more sense in this context. Maybe you could make pull request that cleans this up?

[Jan Blechta]

@garth-wells: Ok.

[Jan Blechta]

Fixes in pull request #126 and pull request #127. If the latter is merged, a deprecation issue should be filed.

[Prof Garth Wells]

Merge into master. See 750b24f, c0710e1 and fd83bbc.