Check that the field solved for is written back to state

[kynan]

When we have a UFL equation for a field Tracer we need to make sure

• the field Tracer is written back to state
• only the field Tracer is written back and no other field

Suppose we have UFL input like this

t = state.scalar_fields["Tracer"]


v = TestFunction(t.element)

f = Coefficient(t.element)



A = dot(grad(v),grad(u))*dx

RHS = v*f*dx



state.scalar_fields["Tracer"]=solve(A, RHS)


we need to check that there is an assignment to state.scalar_fields["Tracer"] but not to any other field in state

[gmarkall]

I'm not sure I understand this issue - isn't this what happens anyway? For example, in the euler-advection.ufl test case, only the Tracer field is returned to the host, not the velocity.

[gmarkall]

Having started to read https://github.com/gmarkall/manycore_form_compiler/wiki/UFL-equations-for-Fluidity-fields , I now see that this applies to the UFL code for solving for a particular field in an flml, and is not a general code-generation issue where we were returning the wrong things back to state.

[gmarkall]

I think i see another issue with this. In tests/inputs/shallow-water.ufl, the final three lines are something like:

(Vnew, Hnew) = split(solve(A,RHS)
state.scalar_fields["Height"] = Hnew
state.scalar_fields["Velocity"] = Vnew

This is done because the pressure and velocity are coupled in this method - how does this fit into the idea that we have to check what the UFL returns when it is the code for a given field?

[kynan]

Good point, I hadn't thought of this case. Presumably, coupled fields will give us more headaches than this. However, I would assume that in the flml file for the shallow water model it would be indicated that these fields are coupled.

[gmarkall]

I'm not sure this is a valid issue, since we stalled the discussion 10 months ago after mentioning that this isn't that straightforward for coupled fields, and I don't think this is much of an important thing to check for anyway. Shall we close it?