Real function space #909
Real function space #909
Conversation
This must be a hole in our test coverage.
Conflicts: firedrake/ffc_interface.py firedrake/solving.py
Conflicts: .travis.yml
| def exterior_facet_boundary_node_map(self, method): | ||
| """":class:`RealFunctionSpace` objects have no exterior facet boundary | ||
| node map.""" | ||
| return None |
There was a problem hiding this comment.
To allow field splitting to work with real spaces (especially inside bigger problems) we need to enable the .dm property correctly. And also the ._ises property. This will need to be special-cased. So you should implement the _dm method that makes this DM. The way I do it is that PyOP2 supplies the DM and the field_ises via the dof_dset of the function space. This is presumably not implemented for GlobalDataSet objects.
| """:class:`RealFunctionSpace` objects have no node set.""" | ||
| self.node_set = None | ||
| """:class:`RealFunctionSpace` objects have no dof set.""" | ||
| self.dof_dset = op2.GlobalDataSet(self.make_dat()) |
There was a problem hiding this comment.
If I make two RealFunctionSpace objects, they currently don't share a dof_dset. This may be a problem.
There was a problem hiding this comment.
It's not clear to me, in the pyop2 implementation, why a global dataset needs to remember the global used to make it (and not just the shape, etc...).
| # Error checking | ||
| for space in spaces: | ||
| if type(space) is impl.FunctionSpace: | ||
| if type(space) in (impl.FunctionSpace, impl.RealFunctionSpace): |
There was a problem hiding this comment.
Why, when RealFunctionSpace inherits FunctionSpace?
Conflicts: firedrake/assemble.py
Codecov Report
@@ Coverage Diff @@
## master #909 +/- ##
=========================================
Coverage ? 89.02%
=========================================
Files ? 52
Lines ? 6615
Branches ? 931
=========================================
Hits ? 5889
Misses ? 569
Partials ? 157 |
Conflicts: .drone.yml .travis.yml
Conflicts: .drone.yml
When the row-space of a 2-form is in R, we were still using index 1 for the columns, really it should be 0.
.travis.yml
Outdated
| - mkdir tmp | ||
| - cd tmp | ||
| - ../scripts/firedrake-install --disable-ssh --minimal-petsc ${SLEPC} --adjoint --slope --install thetis --install gusto ${PACKAGE_MANAGER} | ||
| - ../scripts/firedrake-install --disable-ssh --minimal-petsc ${SLEPC} --adjoint --slope --install thetis --install gusto ${PACKAGE_MANAGER} --package-branch PyOP2 globals_petsc_changes |
There was a problem hiding this comment.
This branch is already merged, right?
firedrake/functionspaceimpl.py
Outdated
| def __init__(self, mesh, element, name): | ||
|
|
||
| self.name = name | ||
| self._index = None |
firedrake/functionspaceimpl.py
Outdated
| self.comm = mesh.comm | ||
|
|
||
| """:class:`RealFunctionSpace` objects have no node set.""" | ||
| self.node_set = None |
There was a problem hiding this comment.
Most of these are argument independent, so I would make them class attributes:
class RealFunctionSpace(FunctionSpace):
finat_element = None
dim = 1
rank = 0
shape = ()
node_set = None
def __init__(self, mesh, element, name):
self._ufl_element = element
self.name = name
self.comm = mesh.comm
self._mesh = mesh
self.dof_dset = op2.GlobalDataSet(self.make_dat())Note as well we now have finat_element, not fiat_element.
firedrake/tsfc_interface.py
Outdated
|
|
||
| a = form.arguments() | ||
| reals = map(lambda x: x.ufl_element().family() == "Real", a) | ||
| if not a or not any(reals): |
There was a problem hiding this comment.
the second condition is subsumed by the first, since any(()) => False.
The long-awaited R-space. Most of the heavy lifting has already gone into PyOP2. This mostly just makes the Firedrake wrappers behave correctly. There is also a new manual section explaining how this works.