Problem with autodiff in multifield

[jfbarthelemy]

This issue follows the one described here as concerns the problem to solve but involves another error.

Now the idea is to impose a nonlinear flux between the top line of the mesh on which is defined a "lineic" variable uᵗ and the whole domain on which is defined the "bulk" variable u. The problem seems to be correctly solved if the jacobian is explicitly provided but an error is raised in case of autodiff resolution.

I attach an example.

test_multi_nonlinear.jl.txt