BUG: recompute tape is failing (interpolate)

[Ig-dolci]

Describe the bug
recompute tape is failing for the following code:

Steps to Reproduce

from firedrake import *
import firedrake.adjoint as fda

mesh = UnitSquareMesh(2, 2)
V = FunctionSpace(mesh, "DG", 0)
u = Function(V)

fda.continue_annotation()
p = fda.AdjFloat(0.)
u.interpolate(p)
J = assemble(u*dx)

fda.stop_annotating()

Jhat = fda.ReducedFunctional(J, fda.Control(p))
J = Jhat(fda.AdjFloat(1.))
print("J: ", J, " u: ", u.block_variable.checkpoint.dat.data_ro)
J = Jhat(fda.AdjFloat(2.))
print("J: ", J, " u: ", u.block_variable.checkpoint.dat.data_ro)

The output is

J:  0.0  u:  [0. 0. 0. 0. 0. 0. 0. 0. 0.]
J:  0.0  u:  [0. 0. 0. 0. 0. 0. 0. 0. 0.]

The expected output is:

J:  1.0  u:  [1. 1. 1. 1. 1. 1. 1. 1. 1.]
J:  2.0  u:  [2. 2. 2. 2. 2. 2. 2. 2. 2.]

It seems to be an issue related to u.interpolate(p); when we modify it to u.assign(p), the result is right.

Environment:

• OS: MacOS
• Python version: 3.13

[connorjward]

Is this related to #3909?

[jrmaddison]

UFL sees an AdjFloat as being just a float, and creates a FloatValue. I think the conversion is happening in as_ufl.

[dham]

Yes, do we not need to use a function in R in these cases?

[Ig-dolci]

UFL sees an AdjFloat as being just a float, and creates a FloatValue. I think the conversion is happening in as_ufl.

I see.

Yes, do we not need to use a function in R in these cases?

The controls are working in R space. A warning (or raising an error) for this case looks to be good.