Use one function in linearization

[1-Bart-1]

Combining the functions f_x!, f_u! and f_d! into one function f_xud! before linearization would reduce the number of function calls needed for linearization, and thus increase performance. This would give one big jacobian, which can be split up into the A, Bu and Bd matrixes.

https://github.com/JuliaControl/ModelPredictiveControl.jl/blob/9e800029610c43f27a9ad915d3ca21797abcbadf/src/model/linearization.jl#L33C1-L35C87

[franckgaga]

I'm not convinced the increased code complexity and reduced clarity is worth the presumably small performance gain here. There are only two function calls to the state update function solver_f! if you don't have measured disturbances (i.e. f_x! and f_u!).

I'm not even sure that's possible, moreover. The solver_f! is differentiated with respect to a distinct variable on each jacobian! call i.e. with respect to $\mathbf{x}$ for $\mathbf{A}$, and with respect to $\mathbf{u}$ for $\mathbf{B_u}$:

The code right now is consistent with the classical linearization theory: one Jacobian computed for each state-space matrix.

[1-Bart-1]

Ok, nevermind then!