Don't call jac_structure if problem has no constraints

[abelsiqueira]

CC @mlubin

Untested because I don't have KNITRO. Follows jump-dev/Ipopt.jl#71, fixing for JuliaOpt/MathProgBase.jl#133.

[yeesian]

@mlubin I don't know what to check for (it seems like the unit test is supposed to be added at the MPB level), but this seems acceptable to me. Okay to merge?

[mlubin]

If it passes rosenbrocktest then you can merge.

[gragusa]

Tested and won't work (even fixing row 85). Although the nonlinear model is loaded correctly, KNITRO still calls eval_jac_g callback --- it seems the numConstr=0 and empty indexes are not the right wy to communicate to KNITRO that the problem is unconstrained. Should probably pass pointer to null somewhere?

[gragusa]

&& should be ?

[sylvainmouret]

I agree

[sylvainmouret]

The remaining issue mentioned by gragusa is due to the fact that Knitro asks for objective gradient and jacobian evaluations through a single callback.
Then, the interface calls the two separate callbacks internally.
To fix this, you need to modify the line 254
kp.eval_jac_g(x,unsafe_wrap(Array,J_,nnzJ))
by
if m > 0
kp.eval_jac_g(x,unsafe_wrap(Array,J_,nnzJ))
end

[huckl3b3rry87]

@sylvainmouret line 254 of what script?

[sylvainmouret]

Sorry, I forgot to mention it. It's in KNITRO.jl.

[sylvainmouret]

This has been implement in #73 so I will close.