Function(space, vector) constructor is ill-defined in parallel

Creating function from an already initialized vector (which may have no knowledge of DOF map, hence not having ghosts) ensures none of the following:

ghosts are additionally prepared
error is raised when ghosts are missing

Following code demonstrates the problem:

from dolfin import *

mesh = UnitCubeMesh(6, 6, 6)
V = FunctionSpace(mesh, "CG", 1)
v = TestFunction(V)
while True:
    x = assemble(v*dx)
    u = Function(V, x)
    print assemble(u*dx)

With debugging-enabled PETSc it raises PETSC_ERR_ARG_OUTOFRANGE 63 but computes random rubish with production builds.

Note: C++ documentation says *Warning: This constructor is intended for internal library use only*. This is apparently not reflected in python version.

Note: The example above also demonstrates that there is left some debugging print statement somewhere in the code resulting in the output like Local size, dim: 165, 343.

Comments (4)