Matrix inverse is broken

[Technici4n]

ForwardDiff: v0.10.38
Julia: v1.11.3

julia> function f(ε)
            A = [1 ε; 2ε 2]
            inv(A) * A
       end
f (generic function with 1 method)

julia> ForwardDiff.derivative(f, 0)
2×2 Matrix{Float64}:
 0.0  0.0
 1.0  0.0

inv(A) * A is an identity, so its derivative should be 0.

On my Julia version, what is happening is that the inv version will see that istriu(A) evaluates to true and pick an optimized path for that - completely ignoring the dual components on the lower diagonal part.

[KristofferC]

This is fixed on the master branch.